1. An image processing device comprising: a decoding unit configured to decode encoded data including an initial difference coefficient that is a difference between a DC coefficient and a predetermined initial value, a replacement difference coefficient that is a difference between the DC coefficient and a (0, 0) coefficient located at the beginning of an 8x8 quantization matrix, and further difference coefficients each being a difference between two adjacent coefficients in a sequence of coefficients of the 8x8 quantization matrix arranged in scan order; an inverse DPCM unit (552) configured to set the DC coefficient by adding the initial value to the initial difference coefficient and to set the 8x8 quantization matrix by setting the (0, 0) coefficient located at the beginning of the 8x8 quantization matrix by adding the DC coefficient to the replacement difference coefficient and by setting the remaining coefficients of the 8x8 quantization matrix by stepwise adding the further difference coefficients to the (0, 0) coefficient, wherein a remaining coefficient at a particular position in scan order is the sum of the (0, 0) coefficient and the difference coefficient up to said particular position; a matrix size transformation unit configured to upconvert the (8, 8) quantization matrix into a 32x32 quantization matrix by performing a nearest neighbor interpolation process on matrix elements of the 8x8 quantization matrix; and a dequantization unit (440) configured to dequantize quantized data obtained by decoding encoded data, using the DC coefficient and the 32x32 quantization matrix.

2. The image processing device as claimed in Claim 1, wherein the decoding unit is configured to decode the encoded data including a syntax in which the replacement difference coefficient and the difference coefficients are collectively included as a difference coefficient group.

3. The image processing device as claimed in Claim 1, wherein 202 the replacement difference coefficient and the difference coefficients are included as the syntax of the encoded data in the order of the replacement difference coefficient and the difference coefficients, and the decoding unit is configured to decode the replacement difference coefficient and the difference coefficients in the order of the replacement difference coefficient and the difference coefficients.

4. The image processing device as claimed in Claim 3, wherein the initial difference value and the difference coefficient group are included as the syntax of the encoded data in the order of the initial difference value and the difference coefficient group, and the decoding unit is configured to decode the initial difference value and the difference coefficient group in the order of the initial difference value and the difference coefficient group.

5. The image processing device as claimed in Claim 4, wherein the initial difference value and the difference coefficient group are included in the encoded data by performing an exponential golomb encoding process, and the decoding unit is configured to perform an exponential golomb decoding process on the initial difference value and the difference coefficient group obtained by performing the exponential golomb encoding in the order of the initial difference value and the difference coefficient group.

6. The image processing device as claimed in Claim 5, further comprising: an inverse orthogonal transform unit (450) configured to inversely transform the transform coefficient data generated by the dequantization unit.

7. The image processing device as claimed in Claim 6, wherein the inverse orthogonal transform unit (450) is configured to inversely transform compared to the transforming performed by a 32x32 transform unit. 203

8. The image processing device as claimed in Claim 1, further comprising: a replacement unit configured to replace a (0, 0) coefficient located at the beginning of the 32x32 quantization matrix set by the matrix size transformation unit with the DC coefficient set by the inverse DPCM unit;

9. The image processing device as claimed in Claim 8, wherein the dequantization unit (440) is configured to dequantize quantized data obtained by decoding encoded data, using the 32x32 quantization matrix in which the (0, 0) coefficient located at the beginning has been replaced with the DC coefficient by the replacement unit.

10. An image processing method comprising: decoding encoded data including an initial difference coefficient that is a difference between a DC coefficient and a predetermined initial value, a replacement difference coefficient that is a difference between the DC coefficient and a (0, 0) coefficient located at the beginning of an 8x8 quantization matrix, and further difference coefficients each being a difference between two adjacent coefficients in a sequence of coefficients of the 8x8 quantization matrix arranged in scan order; setting the DC coefficient by adding the initial value to the initial difference coefficient and setting the 8x8 quantization matrix by setting the (0, 0) coefficient located at the beginning of the 8x8 quantization matrix by adding the DC coefficient to the replacement difference coefficient and by setting the remaining coefficients of the 8x8 quantization matrix by stepwise adding the further difference coefficients to the (0, 0) coefficient, wherein a remaining coefficient at a particular position in scan order is the sum of the (0, 0) coefficient and the difference coefficient up to said particular position; upconverting the (8, 8) quantization matrix into a 32x32 quantization matrix by performing a nearest neighbor interpolation process on matrix elements of the 8x8 quantization matrix; and dequantizing quantized data obtained by decoding encoded data, using the DC coefficient and the 32x32 quantization matrix. 204

11. The image processing method as claimed in Claim 10, wherein said decoding includes a syntax in which the replacement difference coefficient and the difference coefficients are collectively included as a difference coefficient group.

12. The image processing method as claimed in Claim 10, wherein the replacement difference coefficient and the difference coefficients are included as the syntax of the encoded data in the order of the replacement difference coefficient and the difference coefficients, and the replacement difference coefficient and the difference coefficients are decoded in the order of the replacement difference coefficient and the difference coefficients.

13. The image processing method as claimed in Claim 12, wherein the initial difference value and the difference coefficient group are included as the syntax of the encoded data in the order of the initial difference value and the difference coefficient group, and the initial difference value and the difference coefficient group are decoded in the order of the initial difference value and the difference coefficient group.

14. The image processing method as claimed in Claim 13, wherein the initial difference value and the difference coefficient group are included in the encoded data by performing an exponential golomb encoding process, and the method comprises performing an exponential golomb decoding process on the initial difference value and the difference coefficient group obtained by performing the exponential golomb encoding in the order of the initial difference value and the difference coefficient group.

15. The image processing method as claimed in Claim 14, further comprising: inversely transforming the transform coefficient data generated by said dequantization of the quantized data.

16. The image processing method as claimed in Claim 15, wherein 205 the transform coefficient data is inversely transformed compared to the transforming performed by a 32x32 transform unit.

17. The image processing method as claimed in Claim 10, further comprising: replacing a (0, 0) coefficient located at the beginning of the 32x32 quantization matrix set by said upconverting of the (8, 8) quantization matrix with the DC coefficient set by adding the initial value to the initial difference coefficient and to set the 8x8 quantization matrix by setting the (0, 0) coefficient located at the beginning of the 8x8 quantization matrix by adding the DC coefficient to the replacement difference coefficient and by setting the remaining coefficients of the 8x8 quantization matrix by stepwise adding the further difference coefficients to the (0, 0) coefficient.

18. The image processing method as claimed in Claim 17, further comprising: dequantizing quantized data obtained by decoding encoded data, using the 32x32 quantization matrix in which the (0, 0) coefficient located at the beginning has been replaced with the DC coefficient.

Technical Field
[0001]The present disclosure relates to an image processing
device and method.
Background Art
[0002]In H.264/AVC (Advanced Video Coding), which is one of
standard specifications of video coding schemes, the
profiles of High Profile or higher allow quantization of
image data with quantization step sizes that differ from one
component of orthogonal transform coefficient to another.
The quantization step size for each component of orthogonal
transform coefficient may be set based on a reference step
value and a quantization matrix (also referred to as a
scaling list) defined by a size equivalent to the unit of an
orthogonal transform.
[0003]
A specified value of a quantization matrix is prepared
for each prediction mode (intra-prediction mode, interprediction
mode) and for each transform unit size (4x4, 8x8).
Furthermore, users are allowed to specify a unique
quantization matrix different from the specified values in a
sequence parameter set or picture parameter set. In a case
where no quantization matrices are used, quantization step
sizes used for quantization have an equal value for all the
2
components.
[0004]
In HEVC (High Efficiency Video Coding), which is being
standardized as a next-generation video coding scheme and
which is a successor to H.264/AVC, the concept of coding
units (CUs) corresponding to traditional macroblocks has
been introduced (see, for example, NPL 1). The range of
sizes of coding units is specified by a set of values which
are powers of 2, called the largest coding unit (LCU) and
the smallest coding unit (SCU), in a sequence parameter set.
Furthermore, the specific coding unit size in the range
specified by the LCU and the SCU is specified using
splitflag.
[0005]
In HEVC, one coding unit may be divided into one or
more orthogonal transform units, or one or more transform
units (TUs). An available transform unit size is any of 4x4,
8x8, 16x16, and 32x32.
[0006]
Meanwhile, the DC component (also referred to as the
direct current component) of a quantization matrix (scaling
list) is transmitted as data different from the AC
components (also referred to as the alternating current
components) thereof for purposes such as the reduction in
the amount of coding during transmission. Specifically, the
3
DC component of a scaling list is transmitted as a DC
coefficient (also referred to as a direct current
coefficient) different from AC coefficients (also referred
to as alternating current coefficients), which are the AC
components of the scaling list.
[0007]
In order to reduce the amount of coding of the DC
coefficient during transmission, it has been suggested that
a constant (for example, 8) is subtracted from the value of
the DC coefficient and the resulting value
(scaling_list_dc_coef_minus8) is encoded using signed
exponential Golomb coding (see, for example, NPL 1 ) .
Citation List
Non Patent Literature
[0008]
NPL 1: Benjamin Bross, Fraunhofer HHI, Woo-Jin Han,
Gachon University, Jens-Rainer Ohm, RWTH Aachen, Gary J.
Sullivan, Microsoft, Thomas Wiegand, Fraunhofer HHI / TU
Berlin, JCTVC-H1003, "High Efficiency Video Coding (HEVC)
text specification draft 6", Joint Collaborative Team on
Video Coding (JCT-VC) of ITU-T SG16 WP3 and ISO/IEC
JTC1/SC29/WG11 7th Meeting: Geneva, CH, 21-30 November, 2011
Summary of Invention
Technical Problem
[0009]
4
However, there is a concern that the method described
above will not provide sufficient compression efficiency
although it facilitates processes.
[0010]
The present disclosure has been made in view of the
situation described above, and it is an object of the
present disclosure to enable suppression of an increase in
the amount of coding of a scaling list.
Solution to Problem
[0011]
An aspect of the present disclosure provides an image
processing device including a setting unit configured to set
a coefficient located at the beginning of a quantization
matrix whose size is limited to not greater than a
transmission size that is a maximum size allowed in
transmission, by adding a replacement difference coefficient
that is a difference between a replacement coefficient and
the coefficient located at the beginning of the quantization
matrix to the coefficient located at the beginning of the
quantization matrix, the replacement coefficient being used
to replace a coefficient located at the beginning of an upconverted
quantization matrix which is obtained by upconverting
the quantization matrix to the same size as a
block size that is a unit of processing in which
dequantization is performed; an up-conversion unit
5
configured to up-convert the quantization matrix set by the
setting unit to set the up-converted quantization matrix;
and a dequantization unit configured to dequantize quantized
data obtained by decoding encoded data, using an upconverted
quantization matrix in which a coefficient located
at the beginning of the up-converted quantization matrix set
by the up-conversion unit has been replaced with the
replacement coefficient.
[0012]
The setting unit can set the replacement coefficient by
adding a difference between the replacement coefficient and
an initial value set for the quantization matrix to the
initial value.
[0013]
The setting unit can set coefficients of the
quantization matrix using the replacement difference
coefficient and difference coefficients that are differences
between the coefficients of the quantization matrix.
[0014]
The replacement difference coefficient and the
difference coefficients that are the differences between the
coefficients of the quantization matrix can be collectively
transmitted. The setting unit can set the coefficients of
the quantization matrix using the collectively transmitted
replacement difference coefficient and difference
6
coefficients.
[0015]
The replacement difference coefficient and the
difference coefficients that are the differences between the
coefficients of the quantization matrix can have been
encoded. The setting unit can decode the encoded
replacement difference coefficient and the encoded
difference coefficients.
[0016]
The up-conversion unit can up-convert the quantization
matrix whose size is limited to not greater than the
transmission size, by performing a nearest neighbor
interpolation process on matrix elements of the quantization
matrix.
[0017]
The transmission size can be an 8x8 size. The upconversion
unit can up-convert a quantization matrix having
an 8x8 size to a quantization matrix having a 16x16 size, by
performing the nearest neighbor interpolation process on
matrix elements of the quantization matrix having the 8x8
size.
[0018]
The up-conversion unit can up-convert a quantization
matrix having an 8x8 size to a quantization matrix having a
32x32 size, by performing the nearest neighbor interpolation
7
process on matrix elements of the quantization matrix having
the 8x8 size.
[0019]
A coding unit that is a unit of processing in which a
decoding process is performed and a transform unit that is a
unit of processing in which a transform process is performed
can have a layered structure. The image processing device
can further include a decoding unit configured to perform a
decoding process on the encoded data using a unit having a
layered structure to generate the quantized data. The upconversion
unit can up-convert the quantization matrix from
the transmission size to a size of a transform unit that is
a unit of processing in which dequantization is performed.
[0020]
An aspect of the present disclosure provides an image
processing method including setting a coefficient located at
the beginning of a quantization matrix whose size is limited
to not greater than a transmission size that is a maximum
size allowed in transmission, by adding a replacement
difference coefficient that is a difference between a
replacement coefficient and the coefficient located at the
beginning of the quantization matrix to the coefficient
located at the beginning of the quantization matrix, the
replacement coefficient being used to replace a coefficient
located at the beginning of an up-converted quantization
8
matrix which is obtained by up-converting the quantization
matrix to the same size as a block size that is a unit of
processing in which dequantization is performed; upconverting
the set quantization matrix to set the upconverted
quantization matrix; and dequantizing quantized
data obtained by decoding encoded data, using an upconverted
quantization matrix in which a coefficient located
at the beginning of the set up-converted quantization matrix
has been replaced with the replacement coefficient.
[0021]
Another aspect of the present disclosure provides an
image processing device including a setting unit configured
to set a replacement difference coefficient that is a
difference between a replacement coefficient and a
coefficient located at the beginning of a quantization
matrix whose size is limited to not greater than a
transmission size that is a maximum size allowed in
transmission, the replacement coefficient being used to
replace a coefficient located at the beginning of an upconverted
quantization matrix which is obtained by upconverting
the quantization matrix to the same size as a
block size that is a unit of processing in which
dequantization is performed; a quantization unit configured
to quantize an image to generate quantized data; and a
transmission unit configured to transmit encoded data
9
obtained by encoding the quantized data generated by the
quantization unit, replacement coefficient data obtained by
encoding the replacement coefficient, and replacement
difference coefficient data obtained by encoding the
replacement difference coefficient set by the setting unit.
[0022]
The setting unit can set a difference between the
replacement coefficient and an initial value set for the
quantization matrix.
[0023]
The setting unit can set difference coefficients that
are differences between coefficients of the quantization
matrix. The transmission unit can transmit difference
coefficient data obtained by encoding the difference
coefficients set by the setting unit.
[0024]
The transmission unit can collectively transmit the
replacement coefficient data and the replacement difference
coefficient data.
[0025]
The transmission unit can transmit the replacement
coefficient data and the replacement difference coefficient
data in order of the replacement coefficient data and the
replacement difference coefficient data.
[0026]
10
The quantization unit can quantize the image using the
quantization matrix or the up-converted quantization matrix.
[0027]
A coding unit that is a unit of processing in which an
encoding process is performed and a transform unit that is a
unit of processing in which a transform process is performed
can have a layered structure. The image processing device
can further include an encoding unit configured to encode
the quantized data generated by the quantization unit.
[0028]
Another aspect of the present disclosure provides an
image processing method including setting a replacement
difference coefficient that is a difference between a
replacement coefficient and a coefficient located at the
beginning of a quantization matrix whose size is limited to
not greater than a transmission size that is a maximum size
allowed in transmission, the replacement coefficient being
used to replace a coefficient located at the beginning of an
up-converted quantization matrix which is obtained by upconverting
the quantization matrix to the same size as a
block size that is a unit of processing in which
dequantization is performed; quantizing an image to generate
quantized data; and transmitting encoded data obtained by
encoding the generated quantized data, replacement
coefficient data obtained by encoding the replacement
11
coefficient, and replacement difference coefficient data
obtained by encoding the set replacement difference
coefficient.
[0029]
Still another aspect of the present disclosure provides
an image processing device including a decoding unit
configured to decode encoded data to generate quantized
data; and a dequantization unit configured to dequantize the
quantized data generated by the decoding unit, using a
default quantization matrix having the same size as a block
size that is a unit of processing in which dequantization is
performed, when in a copy mode in which a quantization
matrix is copied, quantization matrix reference data
identifying a reference destination of the quantization
matrix matches quantization matrix identification data
identifying the quantization matrix.
[0030]
The dequantization unit can dequantize the quantized
data by parsing syntax whose semantics is set so that the
default quantization matrix is referred to when the
quantization matrix reference data matches the quantization
matrix identification data.
[0031]
The dequantization unit can dequantize the quantized
data by parsing syntax whose semantics is set so that the
12
default quantization matrix is referred to when a difference
between the quantization matrix reference data and the
quantization matrix identification data is equal to 0.
[0032]
Still another aspect of the present disclosure provides
an image processing method including decoding encoded data
to generate quantized data; and dequantizing the quantized
data generated in the decoding, using a default quantization
matrix having the same size as a block size that is a unit
of processing in which dequantization is performed, when in
a copy mode in which a quantization matrix is copied,
quantization matrix reference data identifying a reference
destination of the quantization matrix matches quantization
matrix identification data identifying the quantization
matrix.
[0033]
Still another aspect of the present disclosure provides
an image processing device including an encoding unit
configured to encode an image to generate encoded data; and
a setting unit configured to set, as syntax of the encoded
data generated by the encoding unit, syntax whose semantics
is set so that a default quantization matrix having the same
size as a block size that is a unit of processing in which
quantization is performed is referred to when in a copy mode
in which a quantization matrix is copied, quantization
13
matrix reference data identifying a reference destination of
the quantization matrix matches quantization matrix
identification data identifying the quantization matrix.
[0034]
Still another aspect of present disclosure provides an
image processing method including encoding an image to
generate encoded data; and setting, as syntax of the
generated encoded data, syntax whose semantics is set so
that a default quantization matrix having the same size as a
block size that is a unit of processing in which
quantization is performed is referred to when in a copy mode
in which a quantization matrix is copied, quantization
matrix reference data identifying a reference destination of
the quantization matrix matches quantization matrix
identification data identifying the quantization matrix.
[0035]
In an aspect of the present disclosure, a coefficient
located at the beginning of a quantization matrix whose size
is limited to not greater than a transmission size that is a
maximum size allowed in transmission is set by adding a
replacement difference coefficient that is a difference
between a replacement coefficient and the coefficient
located at the beginning of the quantization matrix to the
coefficient located at the beginning of the quantization
matrix, the replacement coefficient being used to replace a
14
coefficient located at the beginning of an up-converted
quantization matrix which is obtained by up-converting the
quantization matrix to the same size as a block size that is
a unit of processing in which dequantization is performed;
the set quantization matrix is up-converted to set the upconverted
quantization matrix; and quantized data obtained
by decoding encoded data is dequantized using an upconverted
quantization matrix in which a coefficient located
at the beginning of the set up-converted quantization matrix
has been replaced with the replacement coefficient.
[0036]
In another aspect of the present disclosure, a
replacement difference coefficient that is a difference
between a replacement coefficient and a coefficient located
at the beginning of a quantization matrix whose size is
limited to not greater than a transmission size that is a
maximum size allowed in transmission is set, the replacement
coefficient being used to replace a coefficient located at
the beginning of an up-converted quantization matrix which
is obtained by up-converting the quantization matrix to the
same size as a block size that is a unit of processing in
which dequantization is performed; an image is quantized to
generate quantized data; and encoded data obtained by
encoding the generated quantized data, replacement
coefficient data obtained by encoding the replacement
15
coefficient, and replacement difference coefficient data
obtained by encoding the set replacement difference
coefficient are transmitted.
[0037]
In still another aspect of the present disclosure,
encoded data is decoded to generate quantized data; and the
quantized data generated in the decoding is dequantized
using a default quantization matrix having the same size as
a block size that is a unit of processing in which
dequantization is performed, when in a copy mode in which a
quantization matrix is copied, quantization matrix reference
data identifying a reference destination of the quantization
matrix matches quantization matrix identification data
identifying the quantization matrix.
[0038]
In still another aspect of the present disclosure, an
image is encoded to generate encoded data; and syntax whose
semantics is set so that a default quantization matrix
having the same size as a block size that is a unit of
processing in which quantization is performed is referred to
when in a copy mode in which a quantization matrix is copied,
quantization matrix reference data identifying a reference
destination of the quantization matrix matches quantization
matrix identification data identifying the quantization
matrix is set as syntax of the generated encoded data.
16
Advantageous Effects of Invention
[0039]
According to the present disclosure, it is possible to
process an image. In particular, it is possible to suppress
an increase in the amount of coding of a quantization matrix.
Brief Description of Drawings
[0040]
[Fig. 1] Fig. 1 is a diagram illustrating an example of
a scaling list.
[Fig. 2] Fig. 2 is a diagram illustrating an example of
up-conversion.
[Fig. 3] Fig. 3 is a diagram illustrating an example of
how a scaling list is used in a decoder.
[Fig. 4] Fig. 4 is a diagram illustrating an example of
the encoding of a scaling list.
[Fig. 5] Fig. 5 is a diagram illustrating an example of
the encoding of a scaling list using the present technology.
[Fig. 6] Fig. 6 is a diagram illustrating an example of
exponential Golomb codes.
[Fig. 7] Fig. 7 includes diagrams illustrating an
example of the syntax for a scaling list.
[Fig. 8] Fig. 8 is a diagram illustrating an example of
the syntax for a default matrix.
[Fig. 9] Fig. 9 includes diagrams illustrating examples
of the semantics of a default matrix.
17
[Fig. 10] Fig. 10 is a diagram illustrating an example
of the syntax for a scaling list.
[Fig. 11] Fig. 11 is a diagram illustrating an example
of the syntax for a scaling list using the present
technology.
[Fig. 12] Fig. 12 includes diagrams illustrating an
example of the syntax of a scaling list in the related art.
[Fig. 13] Fig. 13 is a diagram illustrating an example
of the syntax of a scaling list.
[Fig. 14] Fig. 14 is a block diagram illustrating an
example of a main configuration of an image encoding device.
[Fig. 15] Fig. 15 is a block diagram illustrating an
example of a main configuration of an orthogonal
transform/quantization unit.
[Fig. 16] Fig. 16 is a block diagram illustrating an
example of a main configuration of a matrix processing unit.
[Fig. 17] Fig. 17 is a diagram illustrating an example
of downsampling.
[Fig. 18] Fig. 18 is a diagram illustrating an example
of the removal of an overlapping portion.
[Fig. 19] Fig. 19 is a block diagram illustrating an
example of a main configuration of a DPCM unit.
[Fig. 20] Fig. 20 is a flowchart illustrating an
example of the flow of a quantization matrix encoding
process.
18
[Fig. 21] Fig. 21 is a flowchart illustrating an
example of the flow of a DPCM process.
[Fig. 22] Fig. 22 is a block diagram illustrating an
example of a main configuration of an image decoding device.
[Fig. 23] Fig. 23 is a block diagram illustrating an
example of a main configuration of a dequantization/inverse
orthogonal transform unit.
[Fig. 24] Fig. 24 is a block diagram illustrating an
example of a main configuration of a matrix generation unit.
[Fig. 25] Fig. 25 is a diagram illustrating an example
of a nearest neighbor interpolation process.
[Fig. 26] Fig. 26 is a block diagram illustrating an
example of a main configuration of an inverse DPCM unit.
[Fig. 27] Fig. 27 is a flowchart illustrating an
example of the flow of a matrix generation process.
[Fig. 28] Fig. 28 is a flowchart illustrating an
example of the flow of a residual signal decoding process.
[Fig. 29] Fig. 29 is a flowchart illustrating an
example of the flow of an inverse DPCM process.
[Fig. 30] Fig. 30 is a diagram illustrating another
example of the syntax of a scaling list.
[Fig. 31] Fig. 31 is a block diagram illustrating
another example configuration of the DPCM unit.
[Fig. 32] Fig. 32 is a flowchart illustrating another
example of the flow of the DPCM process.
19
[Fig. 33] Fig. 33 is a block diagram illustrating
another example configuration of the inverse DPCM unit.
[Fig. 34] Fig. 34 is a flowchart illustrating another
example of the flow of the inverse DPCM process.
[Fig. 35] Fig. 35 is a diagram illustrating still
another example of the syntax of a scaling list.
[Fig. 36] Fig. 36 is a flowchart illustrating still
another example of the flow of the inverse DPCM process.
[Fig. 37] Fig. 37 is a diagram illustrating still
another example of the syntax of a scaling list.
[Fig. 38] Fig. 38 is a block diagram illustrating still
another example configuration of the DPCM unit.
[Fig. 39] Fig. 39 is a flowchart illustrating still
another example of the DPCM process.
[Fig. 40] Fig. 40 is a block diagram illustrating still
another example configuration of the inverse DPCM unit.
[Fig. 41] Fig. 41 is a flowchart illustrating still
another example of the flow of the inverse DPCM process.
[Fig. 42] Fig. 42 is a flowchart continued from Fig. 41,
illustrating still another example of the flow of the
inverse DPCM process.
[Fig. 43] Fig. 43 includes diagrams illustrating still
another example of the syntax of a scaling list.
[Fig. 44] Fig. 44 includes diagrams illustrating still
another example of the syntax of a scaling list.
20
[Fig. 45] Fig. 45 includes diagrams illustrating still
another example of the syntax of a scaling list.
[Fig. 46] Fig. 46 is a diagram illustrating an example
of a multi-view image encoding scheme.
[Fig. 47] Fig. 47 is a diagram illustrating an example
of a main configuration of a multi-view image encoding
device to which the present technology is applied.
[Fig. 48] Fig. 48 is a diagram illustrating an example
of a main configuration of a multi-view image decoding
device to which the present technology is applied.
[Fig. 49] Fig. 49 is a diagram illustrating an example
of a layered image encoding scheme.
[Fig. 50] Fig. 50 is a diagram illustrating an example
of a main configuration of a layered image encoding device
to which the present technology is applied.
[Fig. 51] Fig. 51 is a diagram illustrating an example
of a main configuration of a layered image decoding device
to which the present technology is applied.
[Fig. 52] Fig. 52 is a block diagram illustrating an
example of a main configuration of a computer.
[Fig. 53] Fig. 53 is a block diagram illustrating an
example of a main configuration of a television apparatus.
[Fig. 54] Fig. 54 is a block diagram illustrating an
example of a main configuration of a mobile terminal device.
[Fig. 55] Fig. 55 is a block diagram illustrating an
21
example of a main configuration of a recording/reproducing
apparatus.
[Fig. 56] Fig. 56 is a block diagram illustrating an
example of a main configuration of an imaging apparatus.
[Fig. 57] Fig. 57 is a block diagram illustrating an
example of the use of scalable coding.
[Fig. 58] Fig. 58 is a block diagram illustrating
another example of the use of scalable coding.
[Fig. 59] Fig. 59 is a block diagram illustrating still
another example of the use of scalable coding.
Description of Embodiments
[0041]
Modes for carrying out the present disclosure
(hereinafter referred to as embodiments) will be described
hereinafter. In this regards, the description will be made
in the following order.
1. First embodiment (exemplary application of present
technology)
2. Second embodiment (image encoding device, image
decoding device: first method)
3. Third embodiment (image encoding device, image
decoding device: second method)
4. Fourth embodiment (image encoding device, image
decoding device: third method)
5. Fifth embodiment (image encoding device, image
22
decoding device: fourth method)
6. Sixth embodiment (image encoding device, image
decoding device: other methods)
7. Seventh embodiment (multi-view image encoding device,
multi-view image decoding device)
8. Eighth embodiment (layered image encoding device,
layered image decoding device)
9. Ninth embodiment (computer)
10. Example applications
11. Example applications of scalable coding
[0042]
<1. First Embodiment>
In this embodiment, a description will be given of an
exemplary application of the present technology, which will
be described in detail in the second and following
embodiments thereof.
[0043]
<1-1. Exemplary application of present technology>
First, an exemplary example in which the present
technology is applicable will be described. The present
technology is a technology related to the encoding and
decoding of a scaling list used in quantization and
dequantization processes performed when image data is
encoded and decoded.
[0044]
23
The encoding and decoding of image data may involve
quantization and dequantization of coefficient data. Such
quantization and dequantization are performed in units of a
block having a predetermined size, and a scaling list (or
quantization matrix) having a size corresponding to the
block size is used. For example, in HEVC (High Efficiency
Video Coding), quantization (or dequantization) is performed
with sizes such as 4x4, 8x8, 16x16, and 32x32. In HEVC,
quantization matrices having 4x4 and 8x8 sizes may be
prepared.
[0045]
Fig. 1 illustrates an example of an 8x8 scaling list.
As illustrated in Fig. 1, a scaling list includes a DC
coefficient and AC coefficients. The DC coefficient
composed of one value is the (0, 0) coefficient of a
quantization matrix, and corresponds to the DC coefficient
of a discrete cosine transform (DCT). The AC coefficients
are coefficients of the quantization matrix other than the
(0, 0) coefficient, and correspond to coefficients of the
DCT other than the DC coefficient. Note that, as
illustrated in Fig. 1, the AC coefficients are represented
by a matrix. That is, the AC coefficients also include the
(0, 0) coefficient (hereinafter also referred to as the AC
coefficient (0, 0)), and the (0, 0) coefficient, which is
located at the beginning of the quantization matrix, is
24
replaced with the DC coefficient when used for
quantization/dequantization. Hence, the DC coefficient is
also referred to as a replacement coefficient. In the
example illustrated in Fig. 1, AC coefficients form an 8x8
matrix.
[0046]
In HEVC, furthermore, an up-converted version (upward
conversion) of an 8x8 quantization matrix is used for 16x16
or 32x32 quantization (or dequantization).
[0047]
Fig. 2 illustrates an example of the up-conversion of
an 8x8 scaling list to a 16x16 scaling list. As illustrated
in Fig. 2, a scaling list is up-converted using, for example,
a nearest neighbor interpolation process. The details of
the nearest neighbor interpolation process will be described
below with reference to, for example, Fig. 25. As
illustrated in Fig. 2, up-conversion is performed on the AC
coefficients of the scaling list. Then, the (0, 0)
coefficient among the up-converted AC coefficients is
replaced with the DC coefficient.
[0048]
Two types of 8x8 scaling lists are prepared, namely,
that used for up-conversion to 16x16 ("8x8 for 16x16") and
that used for up-conversion to 32x32 ("8x8 for 32x32").
[0049]
25
The scaling list used for quantization during encoding
(using an encoder) is also used for dequantization during
decoding (using a decoder). That is, the scaling list is
transmitted from the encoding side (the encoder) to the
decoding side (the decoder). Fig. 3 illustrates an example
of the transmission of scaling lists.
[0050]
As in the example illustrated in Fig. 3, the two types
of 8x8 scaling lists, namely, that used for up-conversion to
a 16x16 size and that used for up-conversion to a 32x32 size,
as described above, are transmitted. Although not
illustrated in the drawings, a 4x4 scaling list is also
transmitted.
[0051]
The AC coefficients of the 8x8 scaling list used for
up-conversion to a 16x16 size, which has been transmitted in
the manner described above, are up-converted to the 16x16
size at the decoding side (the decoder) using the nearest
neighbor interpolation process described above, and are used
for the dequantization of a block having a 16x16 size after
the (0, 0) coefficient is replaced with the DC coefficient.
[0052]
Similarly, the AC coefficients of the 8x8 scaling list
used for up-conversion to a 32x32 size, which has been
transmitted in the manner described above, are also up-
26
converted to the 32x32 size at the decoding side (the
decoder) using the nearest neighbor interpolation process
described above, and are used for the dequantization of a
block having a 32x32 size after the (0, 0) coefficient is
replaced with the DC coefficient.
[0053]
<1-2. Encoding of scaling list>
The transmission of scaling lists in the manner
described above will increase the amount of coding
accordingly. Thus, in order to suppress a reduction in
coding efficiency, the scaling lists are encoded using a
certain method to reduce the amount of coding of the scaling
lists. Fig. 4 illustrates an example of the encoding of a
scaling list. Specifically, an 8x8 scaling list is
transmitted as follows.
[0054]
In the case of up-conversion of an 8x8 matrix to a 16x16
matrix:
(1) A difference between the (0, 0) coefficient (that
is, the AC coefficient (0, 0)) of the 8x8 matrix and a
predetermined initial value "8" is taken.
(2) Differences between coefficients (that is, AC
coefficients) (adjacent coefficients in a sequence of
coefficients one-dimensionally arranged in scan order) of
the 8x8 matrix are taken.
27
(3) A difference between the (0, 0) coefficient (that
is, the DC coefficient) of the 16x16 matrix and a
predetermined initial value "8" is taken.
(4) The differences obtained in (1) and (2) and the
difference obtained in (3) are transmitted separately.
[0055]
In the case of up-conversion of an 8x8 matrix to a 32x32
matrix:
(1) A difference between the (0, 0) coefficient (that
is, the AC coefficient (0, 0)) of the 8x8 matrix and a
predetermined initial value "8" is taken.
(2) Differences between coefficients (that is, AC
coefficients) (adjacent coefficients in a sequence of
coefficients one-dimensionally arranged in scan order) of
the 8x8 matrix are taken.
(3) A difference between the (0, 0) coefficient (that
is, the DC coefficient) of the 32x32 matrix and a
predetermined initial value "8" is taken.
(4) The differences obtained in (1) and (2) and the
difference obtained in (3) are transmitted separately.
[0056]
In the method described above, however, the differences
are encoded using signed exponential Golomb coding and are
transmitted in (4). As described above, the difference
obtained in (1) is the difference between the AC coefficient
28
(0, 0) and the initial value "8". Thus, there is a concern
that the amount of coding may be increased if the value of
the AC coefficient (0, 0) is not a value close to the
initial value "8".
[0057]
For example, in Fig. 4, the value of the AC coefficient
(0, 0) is "12", and the value "4" is encoded using signed
exponential Golomb coding and is transmitted as the
difference obtained in (1). That is, 7 bits are required
for the transmission of the difference obtained in (1) and
coding efficiency may be reduced correspondingly. If the
value of the difference obtained in (1) increases, coding
efficiency may further be reduced. The same is true for the
case of an 8x8 scaling list used for up-conversion to a
16x16 size and an 8x8 scaling list used for up-conversion to
a 32x32 size.
[0058]
Meanwhile, the energy of DCT coefficients is generally
concentrated in the DC coefficient and neighboring low-order
coefficients. Therefore, in general, a quantization matrix
also has small values for the DC coefficient and neighboring
coefficients. Furthermore, if values that are significantly
different are used for individual frequencies, a
quantization error may be subjectively noticeable. In order
to suppress such visual deterioration in image quality,
29
consecutive values are used for the DC coefficient and
neighboring coefficients.
[0059]
The (0, 1) coefficient, (1. 0) coefficient, and (1. 1)
coefficient obtained after up-conversion correspond to the
AC coefficient (0, 0) before up-conversion. Furthermore,
the (0, 0) coefficient obtained after up-conversion
corresponds to the DC coefficient.
[0060]
Thus, in scaling lists, the value of the AC coefficient
(0, 0) and the value of the DC coefficient are generally
close to each other. For example, MPEG2, AVC, and HEVC
default matrices take values having such a relationship.
Also in the example illustrated in Fig. 4, the value of the
DC coefficient is the same as the value of AC coefficient (0,
0 ) , that is, "12". Thus, the value of the difference
obtained in (3), that is, the difference between the DC
coefficient and the initial value " 8 " , is also " 4 ".
[0061]
That is, taking a difference between each of the DC
coefficient and the AC coefficient (0, 0 ) , whose values are
close to each other, and the initial value may increase the
difference value therebetween, and may also cause redundancy.
It can be said that there will be a risk of further reducing
coding efficiency.
30
[0062]
To address this, a scaling list is transmitted using
the following method instead of using the method illustrated
in Fig. 4. Fig. 5 illustrates an example of this method.
[0063]
In the case of up-conversion of an 8x8 matrix to a 16x16
matrix:
(1) A difference between the (0, 0) coefficient (that
is, the AC coefficient (0, 0)) of the 8x8 matrix and the (0,
0) coefficient (that is, the DC coefficient) of the 16x16
matrix is taken.
(2) Differences between coefficients (that is, AC
coefficients) (adjacent coefficients in a sequence of
coefficients one-dimensionally arranged in scan order) of
the 8x8 matrix are taken.
(3) A difference between the (0, 0) coefficient (that
is, the DC coefficient) of the 16x16 matrix and a
predetermined initial value "8" is taken.
(4) The differences obtained in (1) to (3) are
collectively transmitted.
[0064]
In the case of up-conversion of an 8x8 matrix to a 32x32
matrix:
(1) A difference between the (0, 0) coefficient (that
is, the AC coefficient (0, 0)) of the 8x8 matrix and the (0,
31
0) coefficient (that is, the DC coefficient) of the 32x32
matrix is taken.
(2) Differences between coefficients (that is, AC
coefficients) (adjacent coefficients in a sequence of
coefficients one-dimensionally arranged in scan order) of
the 8x8 matrix are taken.
(3) A difference between the (0, 0) coefficient (that
is, the DC coefficient) of the 32x32 matrix and a
predetermined initial value "8" is taken.
(4) The differences obtained in (1) to (3) are
collectively transmitted.
[0065]
Similarly to the method illustrated in Fig. 4, in (4),
the differences are encoded using exponential Golomb coding
and are transmitted as exponential Golomb codes.
[0066]
At the destination to which the differences are
transmitted as exponential Golomb codes, when the
exponential Golomb codes are received, the received
exponential Golomb codes are decoded to obtain the
individual differences, and the processes inverse to those
in (1) to (3) described above are performed on the obtained
differences to determine the individual coefficients (the DC
coefficient and the AC coefficients).
[0067]
32
<1-3. Exemplary features of present technology>
Exemplary features of the present technology related to
the transmission method described above will now be
described.
[0068]
<1-3-1. DPCM between AC coefficient (0, 0) and DC
coefficient>
Scaling lists are encoded using differential pulse-code
modulation (DPCM) and are transmitted. In the example
illustrated in Fig. 4, the AC coefficients and the DC
coefficient are DPCM encoded individually, whereas,
according to one of the features of the present technology,
as in the example illustrated in Fig. 5, a difference (also
referred to as a replacement difference coefficient) between
the AC coefficient (0, 0) and the DC coefficient is
determined and transmitted.
[0069]
As described above, the AC coefficient (0, 0) and the
DC coefficient generally take values that are close to each
other. Thus, a difference between the AC coefficient (0, 0)
and the DC coefficient may possibly be smaller than a
difference between the AC coefficient (0, 0) and the initial
value " 8 " . That is, the transmission of a replacement
difference coefficient that is a difference between the AC
coefficient (0, 0) and the DC coefficient using the present
33
technology may be more likely to reduce the amount of coding.
[0070]
For example, in the example illustrated in Fig. 5, the
value of the difference obtained in (1) is "0".
[0071]
Fig. 6 is a table illustrating an example of signed
exponential Golomb coding. As indicated in the table
illustrated in Fig. 6, the exponential Golomb code for the
value "4" has a code length of 7 bits whereas the
exponential Golomb code for the value "0" has a code length
of 1 bit. That is, the method illustrated in Fig. 5 can
reduce the amount of coding by 6 bits compared to the method
illustrated in Fig. 4.
[0072]
In general, a coding amount of approximately 100 bits
to 200 bits is required for the transmission of a
quantization matrix having an 8x8 size. Hence, 6 bits occupy
approximately 6% of the total amount. A reduction in the
amount of coding by 6% in High Level Syntax can be said to
be a very large effect.
[0073]
<1-3-2. Collective transmission of DC coefficient and
AC coefficients>
Fig. 7 illustrates an example of the syntax of a
scaling list. The syntax for the example illustrated in Fig.
34
4 is illustrated in an example illustrated in part A of Fig.
7. Specifically, after the difference between the AC
coefficient (0, 0) and the initial value " 8 " and the
differences between the AC coefficients
(scaling_list_delta_coef) are transmitted, the difference
between the DC coefficient and the initial value " 8"
(scaling_list_dc_coef_minus8) is separately transmitted.
[0074]
In contrast, one of the features of the present
technology is that the difference between the DC coefficient
and the AC coefficient (0, 0) and the differences between
the AC coefficients are arranged in this order and are
collectively transmitted. Specifically, as illustrated in
Fig. 5, after the DC coefficient and the AC coefficients
arranged in a predetermined scan order are one-dimensionally
arranged and the difference between the DC coefficient and
the initial value " 8 " is determined, the differences between
adjacent coefficients in the sequence of coefficients are
determined. Further, the resulting differences (differences
between coefficients) are one-dimensionally arranged in the
order of being obtained and are collectively transmitted.
[0075]
The syntax in this case is illustrated in an example in
part B of Fig. 7. Specifically, initially, the difference
between the DC coefficient and the initial value " 8"
35
(scaling_list_dc_coef_minus8) is transmitted, and then the
difference between the DC coefficient and the AC coefficient
(0, 0) and the differences between the AC coefficients
(scaling_list_delta_coef) are transmitted. That is, the DC
coefficient and the AC coefficients are collectively encoded
and transmitted.
[0076]
In this manner, the collective transmission of the
differences arranged in the order of being obtained allows
the decoding side (the decoder) to which the differences are
transmitted to decode the differences in the order of being
transmitted and obtain the individual coefficients. That is,
a DPCM encoded scaling list can be easily decoded. More
specifically, the processing load can be reduced. In
addition, the rearrangement of the differences is no longer
necessary, resulting in a reduction in buffer capacity.
Furthermore, the respective differences can be decoded in
the order of being supplied, resulting in suppression of an
increase in processing time.
[0077]
<1-3-3. Transmission of default matrix>
Fig. 8 is a diagram illustrating an example of the
syntax for the transmission of a default matrix. In the
related art, as illustrated in Fig. 8, the initial
coefficient (that is, the DC coefficient) is transmitted as
36
"0" to transmit information indicating the use of a default
matrix. That is, the value of the difference between the DC
coefficient and the initial value " 8"
(scaling_list_dc_coef_minus8) is " - 8 " . However, as
illustrated in Fig. 6, the exponential Golomb code for the
value "-8" has a code length of 9 bits. That is, there is a
concern that coding efficiency may be significantly reduced.
In general, it is desirable that the number of bits of High
Level Syntax is as small as possible. In addition, as
illustrated in Fig. 8, due to the increased complexity of
the syntax, the processing load may be increased.
[0078]
To address these issues, the initial coefficient is not
set to " 0 " but the semantics of
scaling_list_pred_matrix_id_delta is modified. More
specifically, the semantics of
scaling_list_pred_matrix_id_delta is modified from that
illustrated in part A of Fig. 9 to that illustrated in part
B of Fig. 9. That is, in the related art, as illustrated in
part A of Fig. 9, the value equal to " 0 " indicates that the
immediately preceding matrix (MatrixID - 1) is referred to.
Instead of this description, as illustrated in part B of Fig.
9, the value of scaling_list_pred_matrix_id_delta equal to
"0" means that a default matrix is referred to.
[0079]
37
Accordingly, the code length of an exponential Golomb
code for the transmission of information indicating the use
of a default matrix can be equal to 1 bit, and a reduction
in coding efficiency can be suppressed. Furthermore, in the
related art, syntax as illustrated in parts A and B of Fig.
10 is necessary for a scaling list. This syntax can be
simplified as in an example illustrated in Fig. 11. That is,
the processing load involved in the encoding and decoding of
a scaling list can be reduced.
[0080]
<1-4. Features of syntax with use of present
technology>
Syntax will be more specifically described.
[0081]
In the example of the related art illustrated in parts
A and B of Fig. 10, the determination of default needs to be
performed twice, namely, scaling_list_dc_coef_minus8 and
scaling_list_delta_coef. In addition, for
scaling_list_delta_coef, determination is made in the middle
of the "for" loop, and the loop exits when
useDefaultScalingMatrixFlag = 1. Furthermore, an
intermediate flag called "stopNow" is needed, and, because
of this condition, a branch such as substituting nextCoef
into the value of scalingList further exists. In this
manner, the syntax of the related art involves complicated
38
processing.
[0082]
In the present technology, accordingly, as in the
example illustrated in Fig. 11, the DC coefficient
calculated from scaling_list_dc_coef_minus8 is substituted
into nextCoef to set the initial value of
scaling_list_delta_coef to the DC coefficient.
[0083]
Furthermore, in semantics, the value of
scaling_list_pred_matrix_id_delta, which is represented by
"+1" in the related art, remains unchanged, and the value
"0" is used as a special value.
[0084]
That is to say, in the related art, when
ScalingList[0][2] is to be decoded (matrixId = 2 ) , if
scaling_list_pred_matrix_id_delta = 0, then matrixId = 2 is
obtained from refMatrixId = matrixId - (1+
scaling_list_pred_matrix_id_delta). Thus, refMatrixId = 1
is obtained, and the value of ScalingList[0][1] is copied.
[0085]
In contrast, in the present technology, refMatrixId =
matrixId - scaling_list_pred_matrix_id_delta is set. When
ScalingList[0][2] is to be decoded (matrixId = 2 ) ,
scaling_list_pred_matrix_id_delta = 1 may be set if
ScalingList[0][1] is to be copied (or if refMatrixId = 1 is
39
to be obtained).
[0086]
Accordingly, as illustrated in Fig. 11, the number of
rows of the syntax for a scaling list can be significantly
reduced. In addition, two variables to be included as
intermediate data, namely, UseDefaultScalingMatrix and
stopNow, can be omitted. Furthermore, branch made in the
"for" loop as illustrated in Fig. 10 can be no longer
required. Therefore, the processing load involved in the
encoding and decoding of a scaling list can be reduced.
[0087]
<1-5. Processing units implementing present
technology>
In a case where the present technology is applied to
the transmission of a scaling list, a scaling list is
encoded and decoded in the manner described above.
Specifically, an image encoding device 10 described below
with reference to Fig. 14 encodes a scaling list and
transmits the encoded scaling list, and an image decoding
device 300 described below with reference to Fig. 22
receives and decodes the encoded scaling list.
[0088]
A scaling list is encoded by a matrix processing unit
150 (Fig. 15) in an orthogonal transform/quantization unit
14 (Fig. 14) of the image encoding device 10. More
40
specifically, a scaling list is encoded by a DPCM unit 192
and an exp-G unit 193 (both are illustrated in Fig. 16) in
an entropy encoding unit 164 (Fig. 16) in the matrix
processing unit 150. That is, the DPCM unit 192 determines
differences between coefficients (the DC coefficient and the
AC coefficients) of the scaling list, and the exp-G unit 193
encodes the individual differences using exponential Golomb
coding.
[0089]
In order to encode a scaling list using the present
technology as described above, the DPCM unit 192 may have an
example configuration as illustrated in, for example, Fig.
19, and may perform a DPCM process as in an example
illustrated in Fig. 21. Furthermore, semantics as in an
example illustrated in part C of Fig. 44 or part C of Fig.
45 may be used.
[0090]
In other words, only the DPCM unit 192 and the exp-G
unit 193 may be required to achieve the encoding of a
scaling list using the present technology, and other
components having any configuration may be used as desired.
A necessary configuration, such as a processing unit for upconverting
a scaling list and a processing unit for
performing quantization using a scaling list, may be
provided in accordance with embodiments.
41
[0091]
Furthermore, a scaling list is decoded by a matrix
generation unit 410 (Fig. 23) in a dequantization/inverse
orthogonal transform unit 313 (Fig. 22) of the image
decoding device 300. More specifically, a scaling list is
decoded by an exp-G unit 551 and an inverse DPCM unit 552
(Fig. 24) in an entropy decoding unit 533 (Fig. 24) in the
matrix generation unit 410. That is, the exp-G unit 551
decodes the Golomb codes to obtain differences, and the
inverse DPCM unit 552 determines individual coefficients
(the DC coefficient and the AC coefficients) of the scaling
list from the respective differences.
[0092]
In order to decode an encoded scaling list using the
present technology as described above, the inverse DPCM unit
552 may have an example configuration as illustrated in, for
example, Fig. 26, and may perform an inverse DPCM process as
in an example illustrated in Fig. 29. Furthermore,
semantics as in an example illustrated in part C of Fig. 44
or part C of Fig. 45 may be used.
[0093]
In other words, only the exp-G unit 551 and the inverse
DPCM unit 552 may be required to achieve the decoding of a
scaling list using the present technology, and other
components having any configuration may be used as desired.
42
A necessary configuration, such as a processing unit for upconverting
a scaling list and a processing unit for
performing dequantization using a scaling list, may be
provided in accordance with embodiments.
[0094]
Individual embodiments to which the present technology
is applied will be described hereinafter for more detailed
description of the present technology.
[0095]
<2. Second Embodiment>
<2-1. Syntax: First method>
(1) Syntax of related art
First, Fig. 12 illustrates an example of the syntax of
a quantization matrix (or scaling list) in the related art.
In actual use, a difference matrix between a scaling list
and a prediction matrix thereof, rather than the scaling
list, is generally transmitted. Thus, in the following
description of syntax and so forth, it is assumed that the
description of a scaling list can also apply to a difference
matrix.
[0096]
Part A of Fig. 12 illustrates the syntax for scaling
list data (scaling list data syntax), and part B of Fig. 12
illustrates the syntax of a scaling list (scaling list
syntax).
43
[0097]
(1-1) Scaling list data syntax
As illustrated in part A of Fig. 12, the syntax for
scaling list data specifies that a flag
(scaling_list_present_flag) indicating whether or not a
scaling list is provided, a flag
(scaling_list_pred_mode_flag) indicating whether or not the
current mode is a copy mode, information
(scaling_list_pred_matrix_id_delta) indicating which scaling
list to refer to in the copy mode, and so forth are read.
[0098]
(1-2) Scaling list syntax
As illustrated in part B of Fig. 12, the syntax of a
scaling list specifies that the DC coefficient from which a
constant (for example, 8) is subtracted
(scaling_list_dc_coef_minus8), a difference value
(scaling_list_delta_coef) between AC coefficients, and so
forth are read and that the DC coefficient and the AC
coefficients are restored.
[0099]
However, there is a concern that the pieces of syntax
described above will not provide sufficient compression
efficiency of the DC coefficient although it facilitates
processes.
[0100]
44
Accordingly, in order to obtain sufficient compression
efficiency of a DC coefficient (also referred to as a direct
current coefficient), which is the coefficient of the DC
component (direct current component), a difference between
the DC coefficient and another coefficient is determined,
and the difference value is transmitted instead of the DC
coefficient. That is, the difference value is information
for calculating the DC coefficient, and, in other words, is
substantially equivalent to the DC coefficient. However,
the difference value is generally smaller than the DC
coefficient. Therefore, the transmission of the difference
value instead of the DC coefficient may result in a
reduction in the amount of coding.
[0101]
In the following description, for convenience of
description, a scaling list (quantization matrix) has an 8x8
size. A specific example of the method for transmitting a
difference between the DC coefficient and another
coefficient, instead of the DC coefficient, described above
will be described hereinafter.
[0102]
(2) Syntax for first method
For example, 65 coefficients may be transmitted using
DPCM (Differential Pulse Code Modulation), where the DC
coefficient is considered as the element located at the
45
beginning of an 8x8 matrix (AC coefficients) (first method).
[0103]
That is, first, a difference between a predetermined
constant and the DC coefficient is calculated, and is used
as the initial coefficient of DPCM data. Then, a difference
between the DC coefficient and the initial AC coefficient is
calculated, and is used as the second coefficient of the
DPCM data. Then, a difference between the initial AC
coefficient and the second AC coefficient is calculated, and
is used as the third coefficient of the DPCM data.
Subsequently, a difference from the immediately preceding AC
coefficient is calculated, and is used as the fourth
coefficient of the DPCM data, and the following coefficients
of the DPCM data are determined in a manner similar to that
described above. The coefficients of DPCM data generated in
the manner described above are sequentially transmitted,
starting from the initial coefficient.
[0104]
Accordingly, compression ratio can be improved when the
values of the (0, 0) coefficient (AC coefficient) of an 8x8
matrix and the DC coefficient are close to each other. By
implementing the first method described above, an image
encoding device can process the DC coefficient in a manner
similar to that of AC coefficients (alternating current
coefficients), which are the coefficients of the AC
46
components (also referred to as the alternating current
components). Note that, in order to implement the first
method described above, an image decoding device to which
the coefficients described above are transmitted needs to
specially handle only the initial coefficient. Specifically,
the image decoding device needs to extract the DC
coefficient from among the AC coefficients.
[0105]
Fig. 13 illustrates the syntax of a scaling list in the
case described above. In the example illustrated in Fig. 13,
65 difference values (scaling_list_delta_coef) between
coefficients are read, and, among coefficients (nextcoef)
determined from the difference values, the coefficient
(nextcoef) located at the beginning is used as the DC
coefficient (scaling_list_dc_coef) while the other
coefficients are used as the AC coefficients
(ScalingList[i]).
[0106]
An image encoding device that implements the syntax for
the first method described above will be described
hereinafter.
[0107]
<2-2. Image encoding device>
Fig. 14 is a block diagram illustrating an example
configuration of an image encoding device 10 according to an
47
embodiment of the present disclosure. The image encoding
device 10 illustrated in Fig. 14 is an image processing
device to which the present technology is applied and that
is configured to encode input image data and output the
encoded image data. Referring to Fig. 14, the image
encoding device 10 includes an A/D (Analogue to Digital)
conversion unit 11 (A/D), a rearrangement buffer 12, a
subtraction unit 13, an orthogonal transform/quantization
unit 14, a lossless encoding unit 16, an accumulation buffer
17, a rate control unit 18, a dequantization unit 21, an
inverse orthogonal transform unit 22, an adder unit 23, a
deblocking filter 24, a frame memory 25, a selector 26, an
intra prediction unit 30, a motion search unit 40, and a
mode selection unit 50.
[0108]
The A/D conversion unit 11 converts an image signal
input in analog form to image data in digital form, and
outputs a digital image data sequence to the rearrangement
buffer 12.
[0109]
The rearrangement buffer 12 rearranges images included
in the image data sequence input from the A/D conversion
unit 11. After rearranging the images in accordance with a
GOP (Group of Pictures) structure for use in an encoding
process, the rearrangement buffer 12 outputs the image data
48
in which the images have been rearranged to the subtraction
unit 13, the intra prediction unit 30, and the motion search
unit 40.
[0110]
The subtraction unit 13 is supplied with the image data
input from the rearrangement buffer 12 and prediction image
data selected by the mode selection unit 50, which will be
described below. The subtraction unit 13 calculates
prediction error data that represents the difference between
the image data input from the rearrangement buffer 12 and
the prediction image data input from the mode selection unit
50, and outputs the calculated prediction error data to the
orthogonal transform/quantization unit 14.
[0111]
The orthogonal transform/quantization unit 14 performs
an orthogonal transform and quantization on the prediction
error data input from the subtraction unit 13, and outputs
quantized transform coefficient data (hereinafter referred
to as quantized data) to the lossless encoding unit 16 and
the dequantization unit 21. The bit rate of the quantized
data output from the orthogonal transform/quantization unit
14 is controlled in accordance with a rate control signal
supplied from the rate control unit 18. A detailed
configuration of the orthogonal transform/quantization unit
14 will further be described below.
49
[0112]
The lossless encoding unit 16 is supplied with the
quantized data input from the orthogonal
transform/quantization unit 14, information for generating a
scaling list (or quantization matrix) on the decoding side,
and information concerning intra prediction or inter
prediction which is selected by the mode selection unit 50.
The information concerning intra prediction may include, for
example, prediction mode information indicating an optimum
intra-prediction mode for each block. Furthermore, the
information concerning inter prediction may include, for
example, prediction mode information for block-by-block
prediction of motion vectors, differential motion vector
information, reference image information, and so forth.
Moreover, the information for generating a scaling list on
the decoding side may include identification information
indicating a maximum size of a scaling list to be
transmitted (or a difference matrix between a scaling list
(quantization matrix) and a prediction matrix thereof).
[0113]
The lossless encoding unit 16 performs a lossless
encoding process on the quantized data to generate an
encoded stream. The lossless encoding performed by the
lossless encoding unit 16 may be, for example, variablelength
encoding, arithmetic encoding, or the like.
50
Furthermore, the lossless encoding unit 16 multiplexes
information for generating a scaling list into the header
(for example, a sequence parameter set and a picture
parameter set) of the encoded stream. The lossless encoding
unit 16 further multiplexes the information concerning intra
prediction or inter prediction described above into the
header of the encoded stream. After that, the lossless
encoding unit 16 outputs the generated encoded stream to the
accumulation buffer 17.
[0114]
The accumulation buffer 17 temporarily accumulates the
encoded stream input from the lossless encoding unit 16,
using a storage medium such as a semiconductor memory.
After that, the accumulation buffer 17 outputs the
accumulated encoded stream at a rate corresponding to the
bandwidth of a transmission path (or an output line from the
image encoding device 10).
[0115]
The rate control unit 18 monitors the accumulation
buffer 17 to check the availability of capacity. The rate
control unit 18 generates a rate control signal in
accordance with the available capacity of the accumulation
buffer 17, and outputs the generated rate control signal to
the orthogonal transform/quantization unit 14. For example,
when the available capacity of the accumulation buffer 17 is
51
low, the rate control unit 18 generates a rate control
signal for reducing the bit rate of the quantized data.
Alternatively, for example, when the available capacity of
the accumulation buffer 17 is sufficiently high, the rate
control unit 18 generates a rate control signal for
increasing the bit rate of the quantized data.
[0116]
The dequantization unit 21 performs a dequantization
process on the quantized data input from the orthogonal
transform/quantization unit 14. After that, the
dequantization unit 21 outputs transform coefficient data
acquired through the dequantization process to the inverse
orthogonal transform unit 22.
[0117]
The inverse orthogonal transform unit 22 performs an
inverse orthogonal transform process on the transform
coefficient data input from the dequantization unit 21 to
restore prediction error data. After that, the inverse
orthogonal transform unit 22 outputs the restored prediction
error data to the adder unit 23.
[0118]
The adder unit 23 adds together the restored prediction
error data input from the inverse orthogonal transform unit
22 and the prediction image data input from the mode
selection unit 50 to generate decoded image data. After
52
that, the adder unit 23 outputs the generated decoded image
data to the deblocking filter 24 and the frame memory 25.
[0119]
The deblocking filter 24 performs a filtering process
for reducing blocking artifacts caused by the encoding of an
image. The deblocking filter 24 filters the decoded image
data input from the adder unit 23 to remove (or at least
reduce) blocking artifacts, and outputs the filtered decoded
image data to the frame memory 25.
[0120]
The frame memory 25 stores the decoded image data input
from the adder unit 23 and the filtered decoded image data
input from the deblocking filter 24, using a storage medium.
[0121]
The selector 26 reads decoded image data to be filtered,
which is used for intra prediction, from the frame memory 25,
and supplies the read decoded image data to the intra
prediction unit 30 as reference image data. The selector 26
further reads filtered decoded image data, which is used for
inter prediction, from the frame memory 25, and supplies the
read decoded image data to the motion search unit 40 as
reference image data.
[0122]
The intra prediction unit 30 performs an intra
prediction process in each intra-prediction mode on the
53
basis of the image data to be encoded, which is input from
the rearrangement buffer 12, and the decoded image data
supplied via the selector 26. For example, the intra
prediction unit 30 evaluates a prediction result obtained in
each intra-prediction mode using a predetermined cost
function. Then, the intra prediction unit 30 selects an
intra-prediction mode that minimizes the cost function value,
that is, an intra-prediction mode that provides the highest
compression ratio, as an optimum intra-prediction mode.
Furthermore, the intra prediction unit 30 outputs prediction
mode information indicating the optimum intra-prediction
mode, prediction image data, and information concerning
intra prediction, such as the cost function value, to the
mode selection unit 50.
[0123]
The motion search unit 40 performs an inter prediction
process (or an inter-frame prediction process) on the basis
of the image data to be encoded, which is input from the
rearrangement buffer 12, and the decoded image data supplied
via the selector 26. For example, the motion search unit 40
evaluates a prediction result obtained in each prediction
mode using a predetermined cost function. Then, the motion
search unit 40 selects a prediction mode that minimizes the
cost function value, that is, a prediction mode that
provides the highest compression ratio, as an optimum
54
prediction mode. Furthermore, the motion search unit 40
generates prediction image data in accordance with the
optimum prediction mode. The motion search unit 40 outputs
information concerning inter prediction which includes
prediction mode information indicating the selected optimum
prediction mode, the prediction image data, and information
concerning inter prediction, such as the cost function value,
to the mode selection unit 50.
[0124]
The mode selection unit 50 compares the cost function
value for intra prediction, which is input from the intra
prediction unit 30, with the cost function value for inter
prediction, which is input from the motion search unit 40.
Then, the mode selection unit 50 selects a prediction
technique having the smaller one of the cost function values
for intra prediction and inter prediction. If intra
prediction is selected, the mode selection unit 50 outputs
the information concerning intra prediction to the lossless
encoding unit 16, and also outputs the prediction image data
to the subtraction unit 13 and the adder unit 23.
Alternatively, if inter prediction is selected, the mode
selection unit 50 outputs the information concerning inter
prediction described above to the lossless encoding unit 16,
and also outputs the prediction image data to the
subtraction unit 13 and the adder unit 23.
55
[0125]
<2-3. Example configuration of orthogonal
transform/quantization unit>
Fig. 15 is a block diagram illustrating an example of a
detailed configuration of the orthogonal
transform/quantization unit 14 of the image encoding device
10 illustrated in Fig. 14. Referring to Fig. 15, the
orthogonal transform/quantization unit 14 includes a
selection unit 110, an orthogonal transform unit 120, a
quantization unit 130, a scaling list buffer 140, and a
matrix processing unit 150.
[0126]
(1) Selection unit
The selection unit 110 selects a transform unit (TU) to
be used for the orthogonal transform of image data to be
encoded from among a plurality of transform units having
different sizes. Examples of possible sizes of transform
units selectable by the selection unit 110 include 4x4 and
8x8 for H.264/AVC (Advanced Video Coding), and include 4x4,
8x8, 16x16, and 32x32 for HEVC (High Efficiency Video
Coding). The selection unit 110 may select a transform unit
in accordance with, for example, the size or quality of an
image to be encoded, the performance of the image encoding
device 10, or the like. The selection of a transform unit
by the selection unit 110 may be hand-tuned by a user who
56
develops the image encoding device 10. After that, the
selection unit 110 outputs information that specifies the
size of the selected transform unit to the orthogonal
transform unit 120, the quantization unit 130, the lossless
encoding unit 16, and the dequantization unit 21.
[0127]
(2) Orthogonal transform unit
The orthogonal transform unit 120 performs an
orthogonal transform on the image data (that is, prediction
error data) supplied from the subtraction unit 13, in units
of the transform unit selected by the selection unit 110.
The orthogonal transform performed by the orthogonal
transform unit 120 may be, for example, discrete cosine
transform (DCT), Karhunen-Loève transform, or the like.
After that, the orthogonal transform unit 120 outputs
transform coefficient data acquired through the orthogonal
transform process to the quantization unit 130.
[0128]
(3) Quantization unit
The quantization unit 130 quantizes the transform
coefficient data generated by the orthogonal transform unit
120, by using a scaling list corresponding to the transform
unit selected by the selection unit 110. Furthermore, the
quantization unit 130 switches the quantization step size in
accordance with the rate control signal supplied from the
57
rate control unit 18 to change the bit rate of the quantized
data to be output.
[0129]
Furthermore, the quantization unit 130 causes sets of
scaling lists respectively corresponding to a plurality of
transform units selectable by the selection unit 110 to be
stored in the scaling list buffer 140. For example, as in
HEVC, if there are four possible sizes of transform units,
namely, 4x4, 8x8, 16x16, and 32x32, four sets of scaling
lists respectively corresponding to the four sizes may be
stored in the scaling list buffer 140. Note that if a
specified scaling list is used for a given size, only a flag
indicating that the specified scaling list is used (a
scaling list defined by the user is not used) may be stored
in the scaling list buffer 140 in association with the given
size.
[0130]
A set of scaling lists that may be used by the
quantization unit 130 may be typically set for each sequence
of the encoded stream. In addition, the quantization unit
130 may update a set of scaling lists that is set for each
sequence on a picture-by-picture basis. Information for
controlling the setting and update of a set of scaling lists
may be inserted in, for example, a sequence parameter set
and a picture parameter set.
58
[0131]
(4) Scaling list buffer
The scaling list buffer 140 temporarily stores a set of
scaling lists respectively corresponding to a plurality of
transform units selectable by the selection unit 110, using
a storage medium such as a semiconductor memory. The set of
scaling lists stored in the scaling list buffer 140 is
referred to when the matrix processing unit 150 performs a
process described below.
[0132]
(5) Matrix processing unit
The matrix processing unit 150 encodes a scaling list
to be used for encoding (quantization). After that, the
encoded data of the scaling list (hereinafter referred to as
encoded scaling list data) generated by the matrix
processing unit 150 is output to the lossless encoding unit
16, and may be inserted into the header of the encoded
stream.
[0133]
<2-4. Detailed example configuration of matrix
processing unit>
Fig. 16 is a block diagram illustrating an example of a
more detailed configuration of the matrix processing unit
150. Referring to Fig. 16, the matrix processing unit 150
includes a prediction unit 161, a difference matrix
59
generation unit 162, a difference matrix size transformation
unit 163, an entropy encoding unit 164, a decoding unit 165,
and an output unit 166.
[0134]
(1) Prediction unit
The prediction unit 161 generates a prediction matrix.
As illustrated in Fig. 16, the prediction unit 161 includes
a copy unit 171 and a prediction matrix generation unit 172.
[0135]
In a copy mode, the copy unit 171 copies a previously
transmitted scaling list, and uses the copied quantization
matrix as a prediction matrix (or predicts a scaling list of
an orthogonal transform unit to be processed). More
specifically, the copy unit 171 acquires the size and list
ID (ListID) of a previously transmitted scaling list from a
storage unit 202 in the decoding unit 165. The size is
information indicating the size of the scaling list (ranging
from, for example, 4x4 to 32x32). The list ID is information
indicating the type of prediction error data to be quantized.
[0136]
For example, the list ID includes identification
information indicating that the prediction error data to be
quantized is prediction error data (Intra Luma) of the
luminance component which is generated using a prediction
image subjected to intra prediction, prediction error data
60
(Intra Cr) of the color difference component (Cr) which is
generated using a prediction image subjected to intra
prediction, prediction error data (Intra Cb) of the color
difference component (Cb) which is generated using a
prediction image subjected to intra prediction, or
prediction error data (Inter Luma) of the luminance
component which is generated using a prediction image
subjected to inter prediction.
[0137]
The copy unit 171 selects, as a scaling list to be
copied, a previously transmitted scaling list of the same
size as the scaling list (scaling list of an orthogonal
transform unit to be processed) input to the matrix
processing unit 150, and supplies the list ID of the scaling
list to be copied to the output unit 166 to output the list
ID to devices outside the matrix processing unit 150 (the
lossless encoding unit 16 and the dequantization unit 21).
That is, in this case, only the list ID is transmitted to
the decoding side (or is included in encoded data) as
information indicating a prediction matrix generated by
copying the previously transmitted scaling list. Thus, the
image encoding device 10 can suppress an increase in the
amount of coding of a scaling list.
[0138]
Furthermore, in a normal mode, the prediction matrix
61
generation unit 172 acquires a previously transmitted
scaling list from the storage unit 202 in the decoding unit
165, and generates a prediction matrix using the scaling
list (or predicts a scaling list of an orthogonal transform
unit to be processed). The prediction matrix generation
unit 172 supplies the generated prediction matrix to the
difference matrix generation unit 162.
[0139]
(2) Difference matrix generation unit
The difference matrix generation unit 162 generates a
difference matrix (residual matrix) that is a difference
between the prediction matrix supplied from the prediction
unit 161 (the prediction matrix generation unit 172) and the
scaling list input to the matrix processing unit 150. As
illustrated in Fig. 16, the difference matrix generation
unit 162 includes a prediction matrix size transformation
unit 181, a computation unit 182, and a quantization unit
183.
[0140]
The prediction matrix size transformation unit 181
transforms (hereinafter also referred to as converts) the
size of the prediction matrix supplied from the prediction
matrix generation unit 172 so that the size of the
prediction matrix matches the size of the scaling list input
to the matrix processing unit 150.
62
[0141]
For example, if the size of the prediction matrix is
larger than the size of the scaling list, the prediction
matrix size transformation unit 181 downward converts
(hereinafter also referred to as down-converts) the
prediction matrix. More specifically, for example, when the
prediction matrix has a 16x16 size and the scaling list has
an 8x8 size, the prediction matrix size transformation unit
181 down-converts the prediction matrix to an 8x8 prediction
matrix. Note that any method for down-conversion may be
used. For example, the prediction matrix size
transformation unit 181 may reduce the number of elements in
the prediction matrix (hereinafter also referred to as
downsampling) by using a filter (through computation).
Alternatively, the prediction matrix size transformation
unit 181 may also reduce the number of elements in the
prediction matrix by, for example, as illustrated in Fig. 17,
thinning out some of the elements (for example, only the
even numbered elements (in Fig. 17, the elements in solid
black) among the two-dimensional elements) without using a
filter (hereinafter also referred to as subsampling).
[0142]
Furthermore, for example, if the size of the prediction
matrix is smaller than the size of the scaling list, the
prediction matrix size transformation unit 181 upward
63
converts (hereinafter also referred to as up-converts) the
prediction matrix. More specifically, for example, when the
prediction matrix has an 8x8 size and the scaling list has a
16x16 size, the prediction matrix size transformation unit
181 up-converts the prediction matrix to a 16x16 prediction
matrix. Note that any method for up-conversion may be used.
For example, the prediction matrix size transformation unit
181 may increase the number of elements in the prediction
matrix (hereinafter also referred to as upsampling) by using
a filter (through computation). Alternatively, the
prediction matrix size transformation unit 181 may also
increase the number of elements in the prediction matrix by,
for example, copying the individual elements in the
prediction matrix without using a filter (hereinafter also
referred to as inverse subsampling).
[0143]
The prediction matrix size transformation unit 181
supplies the prediction matrix whose size has been made to
match that of the scaling list to the computation unit 182.
[0144]
The computation unit 182 subtracts the scaling list
input to the matrix processing unit 150 from the prediction
matrix supplied from the prediction matrix size
transformation unit 181, and generates a difference matrix
(residual matrix). The computation unit 182 supplies the
64
calculated difference matrix to the quantization unit 183.
[0145]
The quantization unit 183 quantizes the difference
matrix supplied from the computation unit 182. The
quantization unit 183 supplies the quantized difference
matrix to the difference matrix size transformation unit 163.
The quantization unit 183 further supplies information used
for quantization, such as quantization parameters, to the
output unit 166 to output the information to devices outside
the matrix processing unit 150 (the lossless encoding unit
16 and the dequantization unit 21). Note that the
quantization unit 183 may be omitted (that is, the
quantization of the difference matrix may not necessarily be
performed).
[0146]
(3) Difference matrix size transformation unit
The difference matrix size transformation unit 163
converts the size of the difference matrix (quantized data)
supplied from the difference matrix generation unit 162 (the
quantization unit 183) to a size less than or equal to a
maximum size allowed in transmission (hereinafter also
referred to as a transmission size), if necessary. The
maximum size may have any optional value, and is, for
example, 8x8.
[0147]
65
The encoded data output from the image encoding device
10 is transmitted to an image decoding device corresponding
to the image encoding device 10 via, for example, a
transmission path or a storage medium, and is decoded by the
image decoding device. The upper limit of the size (maximum
size) of the difference matrix (quantized data) during such
transmission, or in the encoded data output from the image
encoding device 10, is set in the image encoding device 10.
[0148]
If the size of the difference matrix is larger than the
maximum size, the difference matrix size transformation unit
163 down-converts the difference matrix so that the size of
the difference matrix becomes less than or equal to the
maximum size.
[0149]
Note that, similarly to the down-conversion of the
prediction matrix described above, the difference matrix may
be down-converted using any method. For example,
downsampling may be performed using a filter or the like, or
subsampling which involves thinning out elements may be
performed.
[0150]
Furthermore, the down-converted difference matrix may
have any size smaller than the maximum size. However, in
general, the larger the difference in size between before
66
and after conversion is, the larger the error becomes. It
is thus desirable that the difference matrix be downconverted
to the maximum size.
[0151]
The difference matrix size transformation unit 163
supplies the down-converted difference matrix to the entropy
encoding unit 164. Note that if the size of the difference
matrix is smaller than the maximum size, the down-conversion
described above is not necessary, and therefore the
difference matrix size transformation unit 163 supplies the
difference matrix input thereto to the entropy encoding unit
164 as it is (that is, the down-conversion of the difference
matrix is omitted).
[0152]
(4) Entropy encoding unit
The entropy encoding unit 164 encodes the difference
matrix (quantized data) supplied from the difference matrix
size transformation unit 163 using a predetermined method.
As illustrated in Fig. 16, the entropy encoding unit 164
includes an overlap determination unit (135-degree unit) 191,
a DPCM (Differential Pulse Code Modulation) unit 192, and an
exp-G unit 193.
[0153]
The overlap determination unit 191 determines symmetry
of the difference matrix supplied from the difference matrix
67
size transformation unit 163. If the residue (difference
matrix) represents a 135-degree symmetric matrix, for
example, as illustrated in Fig. 18, the overlap
determination unit 191 removes the data (matrix elements) of
the symmetric part that is overlapping data. If the residue
does not represent a 135-degree symmetric matrix, the
overlap determination unit 191 omits the removal of the data
(matrix elements). The overlap determination unit 191
supplies the data of the difference matrix from which the
symmetric part has been removed, if necessary, to the DPCM
unit 192.
[0154]
The DPCM unit 192 performs DPCM encoding of the data of
the difference matrix from which the symmetric part has been
removed, if necessary, which is supplied from the overlap
determination unit 191, and generates DPCM data. The DPCM
unit 192 supplies the generated DPCM data to the exp-G unit
193.
[0155]
The exp-G unit 193 encodes the DPCM data supplied from
the DPCM unit 192 using signed or unsigned exponential
Golomb codes (hereinafter also referred to as exponential
Golomb codes). The exp-G unit 193 supplies the encoding
result to the decoding unit 165 and the output unit 166.
[0156]
68
(5) Decoding unit
The decoding unit 165 restores a scaling list from the
data supplied from the exp-G unit 193. The decoding unit
165 supplies information concerning the restored scaling
list to the prediction unit 161 as a previously transmitted
scaling list.
[0157]
As illustrated in Fig. 16, the decoding unit 165
includes a scaling list restoration unit 201 and the storage
unit 202.
[0158]
The scaling list restoration unit 201 decodes the
exponential Golomb codes supplied from the entropy encoding
unit 164 (the exp-G unit 193) to restore a scaling list to
be input to the matrix processing unit 150. For example,
the scaling list restoration unit 201 decodes the
exponential Golomb codes using the method corresponding to
the encoding method for the entropy encoding unit 164, and
obtains a difference matrix by performing transformation
opposite to size transformation performed by the difference
matrix size transformation unit 163 and performing
dequantization corresponding to quantization performed by
the quantization unit 183. The scaling list restoration
unit 201 further subtracts the obtained difference matrix
from the prediction matrix to restore a scaling list.
69
[0159]
The scaling list restoration unit 201 supplies the
restored scaling list to the storage unit 202 for storage in
association with the size and the list ID of the scaling
list.
[0160]
The storage unit 202 stores information concerning the
scaling list supplied from the scaling list restoration unit
201. The information concerning the scaling list stored in
the storage unit 202 is used to generate prediction matrices
of other orthogonal transform units which are processed
later in time. That is, the storage unit 202 supplies the
stored information concerning the scaling list to the
prediction unit 161 as information concerning a previously
transmitted scaling list.
[0161]
Note that, instead of storing the information
concerning the scaling list restored in the way described
above, the storage unit 202 may store the scaling list input
to the matrix processing unit 150 in association with the
size and the list ID of the input scaling list. In this
case, the scaling list restoration unit 201 can be omitted.
[0162]
(6) Output unit
The output unit 166 outputs the supplied various types
70
of information to devices outside the matrix processing unit
150. For example, in the copy mode, the output unit 166
supplies the list ID of the prediction matrix supplied from
the copy unit 171 to the lossless encoding unit 16 and the
dequantization unit 21. Furthermore, for example, in the
normal mode, the output unit 166 supplies the exponential
Golomb codes supplied from the exp-G unit 193 and the
quantization parameters supplied from the quantization unit
183 to the lossless encoding unit 16 and the dequantization
unit 21.
[0163]
The output unit 166 further supplies identification
information indicating a maximum size (transmission size)
allowed in the transmission of a scaling list (or a
difference matrix between a scaling list and a prediction
matrix thereof) to the lossless encoding unit 16 as
information for generating a scaling list on the decoding
side. As described above, the lossless encoding unit 16
creates an encoded stream including the information for
generating a scaling list, and supplies the encoded stream
to the decoding side. The identification information
indicating the transmission size may be specified in advance
by level, profile, and the like. In this case, information
concerning the transmission size is shared in advance
between the apparatus on the encoding side and the apparatus
71
on the decoding side. Thus, the transmission of the
identification information described above can be omitted.
[0164]
<2-5. Detailed example configuration of DPCM unit>
Fig. 19 is a block diagram illustrating an example of a
more detailed configuration of the DPCM unit 192. Referring
to Fig. 19, the DPCM unit 192 includes a DC coefficient
encoding unit 211 and an AC coefficient DPCM unit 212.
[0165]
The DC coefficient encoding unit 211 acquires the DC
coefficient from among the coefficients supplied from the
overlap determination unit 191, subtracts the value of the
DC coefficient from a predetermined initial value (for
example, 8) to determine a difference value, and uses the
difference value as the initial (i = 0) difference value
(scaling_list_delta_coef). The DC coefficient encoding unit
211 supplies the calculated difference value
(scaling_list_delta_coef (i = 0)) to the exp-G unit 193 as
the initial coefficient of the scaling list corresponding to
the region of interest being processed.
[0166]
The AC coefficient DPCM unit 212 acquires an AC
coefficient from among the coefficients supplied from the
overlap determination unit 191, and subtracts the value of
the AC coefficient from the immediately previously processed
72
coefficient to determine a difference value
(scaling_list_delta_coef (i > 0)). The AC coefficient DPCM
unit 212 supplies the determined difference value
(scaling_list_delta_coef (i > 0)) to the exp-G unit 193 as a
coefficient of the scaling list corresponding to the region
of interest being processed. Note that when i = 1, the
immediately preceding coefficient is represented by i = 0.
Thus, the "DC coefficient" is the immediately previously
processed coefficient.
[0167]
In this way, the DPCM unit 192 can transmit the DC
coefficient as the element located at the beginning of the
scaling list (AC coefficients). Accordingly, the coding
efficiency of the scaling list can be improved.
[0168]
<2-6. Flow of quantization matrix encoding process>
Next, an example of the flow of a quantization matrix
encoding process executed by the matrix processing unit 150
illustrated in Fig. 16 will be described with reference to a
flowchart illustrated in Fig. 20.
[0169]
When the quantization matrix encoding process is
started, in step S101, the prediction unit 161 acquires a
scaling list (or quantization matrix) for a current region
(also referred to as a region of interest) that is an
73
orthogonal transform unit to be processed.
[0170]
In step S102, the prediction unit 161 determines
whether or not the current mode is the copy mode. If it is
determined that the current mode is not the copy mode, the
prediction unit 161 advances the process to step S103.
[0171]
In step S103, the prediction matrix generation unit 172
acquires a previously transmitted scaling list from the
storage unit 202, and generates a prediction matrix using
the scaling list.
[0172]
In step S104, the prediction matrix size transformation
unit 181 determines whether or not the size of the
prediction matrix generated in step S103 is different from
that of the scaling list for the current region (region of
interest) acquired in step S101. If it is determined that
both sizes are different, the prediction matrix size
transformation unit 181 advances the process to step S105.
[0173]
In step S105, the prediction matrix size transformation
unit 181 converts the size of the prediction matrix
generated in step S103 to the size of the scaling list for
the current region acquired in step S101.
[0174]
74
When the processing of step S105 is completed, the
prediction matrix size transformation unit 181 advances the
process to step S106. If it is determined in step S104 that
the size of the prediction matrix is the same as the size of
the scaling list, the prediction matrix size transformation
unit 181 advances the process to step S106 while skipping
the processing of step S105 (or without performing the
processing of step S105).
[0175]
In step S106, the computation unit 182 subtracts the
scaling list from the prediction matrix to calculate a
difference matrix between the prediction matrix and the
scaling list.
[0176]
In step S107, the quantization unit 183 quantizes the
difference matrix generated in step S106. Note that this
processing may be omitted.
[0177]
In step S108, the difference matrix size transformation
unit 163 determines whether or not the size of the quantized
difference matrix is larger than the transmission size (the
maximum size allowed in transmission). If it is determined
that the size of the quantized difference matrix is larger
than the transmission size, the difference matrix size
transformation unit 163 advances the process to step S109,
75
and down-converts the difference matrix to the transmission
size or less.
[0178]
When the processing of step S109 is completed, the
difference matrix size transformation unit 163 advances the
process to step S110. Furthermore, if it is determined in
step S108 that the size of the quantized difference matrix
is less than or equal to the transmission size, the
difference matrix size transformation unit 163 advances the
process to step S110 while skipping the processing of step
S109 (or without performing the processing of step S109).
[0179]
In step S110, the overlap determination unit 191
determines whether or not the quantized difference matrix
has 135-degree symmetry. If it is determined that the
quantized difference matrix has 135-degree symmetry, the
overlap determination unit 191 advances the process to step
S111.
[0180]
In step S111, the overlap determination unit 191
removes the overlapping portion (overlapping data) in the
quantized difference matrix. After the overlapping data is
removed, the overlap determination unit 191 advances the
process to step S112.
[0181]
76
Furthermore, if it is determined in step S110 that the
quantized difference matrix does not have 135-degree
symmetry, the overlap determination unit 191 advances the
process to step S112 while skipping the processing of step
S111 (or without performing the processing of step S111).
[0182]
In step S112, the DPCM unit 192 performs DPCM encoding
of the difference matrix from which the overlapping portion
has been removed, if necessary.
[0183]
In step S113, the exp-G unit 193 determines whether or
not DPCM data generated in step S112 has a positive or
negative sign. If it is determined that a sign is included,
the exp-G unit 193 advances the process to step S114.
[0184]
In step S114, the exp-G unit 193 encodes the DPCM data
using signed exponential Golomb coding. The output unit 166
outputs generated exponential Golomb codes to the lossless
encoding unit 16 and the dequantization unit 21. When the
processing of step S114 is completed, the exp-G unit 193
advances the process to step S116.
[0185]
Furthermore, if it is determined in step S113 that no
sign is included, the exp-G unit 193 advances the process to
step S115.
77
[0186]
In step S115, the exp-G unit 193 encodes the DPCM data
using unsigned exponential Golomb coding. The output unit
166 outputs generated exponential Golomb codes to the
lossless encoding unit 16 and the dequantization unit 21.
When the processing of step S115 is completed, the exp-G
unit 193 advances the process to step S116.
[0187]
Furthermore, if it is determined in step S102 that the
current mode is the copy mode, the copy unit 171 copies a
previously transmitted scaling list, and uses the copied
scaling list as a prediction matrix. The output unit 166
outputs the list ID corresponding to the prediction matrix
to the lossless encoding unit 16 and the dequantization unit
21 as information indicating the prediction matrix. Then,
the copy unit 171 advances the process to step S116.
[0188]
In step S116, the scaling list restoration unit 201
restores a scaling list. In step S117, the storage unit 202
stores the scaling list restored in step S116.
[0189]
When the processing of step S117 is completed, the
matrix processing unit 150 ends the quantization matrix
encoding process.
[0190]
78
<2-7. Flow of DPCM process>
Next, an example of a flow of the DPCM process executed
in step S112 in Fig. 20 will be described with reference to
a flowchart illustrated in Fig. 21.
[0191]
When the DPCM process is started, in step S131, the DC
coefficient encoding unit 211 determines a difference
between the DC coefficient and a constant. In step S132,
the AC coefficient DPCM unit 212 determines a difference
between the DC coefficient and the initial AC coefficient.
[0192]
In step S133, the AC coefficient DPCM unit 212
determines whether or not all the AC coefficients have been
processed. If it is determined that there is an unprocessed
AC coefficient, the AC coefficient DPCM unit 212 advances
the process to step S134.
[0193]
In step S134, the AC coefficient DPCM unit 212 shifts
the processing target to the subsequent AC coefficient. In
step S135, the AC coefficient DPCM unit 212 determines a
difference between the previously processed AC coefficient
and the current AC coefficient being processed. When the
processing of step S135 is completed, the AC coefficient
DPCM unit 212 returns the process to step S133.
[0194]
79
In this manner, as long as it is determined in step
S133 that there is an unprocessed AC coefficient, the AC
coefficient DPCM unit 212 repeatedly executes the processing
of steps S133 to S135. If it is determined in step S133
that there is no unprocessed AC coefficient, the AC
coefficient DPCM unit 212 ends the DPCM process, and returns
the process to Fig. 20.
[0195]
As described above, a difference between the DC
coefficient and the AC coefficient located at the beginning
among the AC coefficients is determined, and the difference
instead of the DC coefficient is transmitted to an image
decoding device. Thus, the image encoding device 10 can
suppress an increase in the amount of coding of a scaling
list.
[0196]
Next, an example configuration of an image decoding
device according to an embodiment of the present disclosure
will be described.
[0197]
<2-8. Image decoding device>
Fig. 22 is a block diagram illustrating an example
configuration of an image decoding device 300 according to
an embodiment of the present disclosure. The image decoding
device 300 illustrated in Fig. 22 is an image processing
80
device to which the present technology is applied and that
is configured to decode encoded data generated by the image
encoding device 10. Referring to Fig. 22, the image
decoding device 300 includes an accumulation buffer 311, a
lossless decoding unit 312, a dequantization/inverse
orthogonal transform unit 313, an adder unit 315, a
deblocking filter 316, a rearrangement buffer 317, a D/A
(Digital to Analogue) conversion unit 318, a frame memory
319, selectors 320 and 321, an intra prediction unit 330,
and a motion compensation unit 340.
[0198]
The accumulation buffer 311 temporarily accumulates an
encoded stream input via a transmission path, using a
storage medium.
[0199]
The lossless decoding unit 312 decodes the encoded
stream input from the accumulation buffer 311 in accordance
with the encoding scheme used for encoding. The lossless
decoding unit 312 further decodes the information
multiplexed in the header region of the encoded stream. The
information multiplexed in the header region of the encoded
stream may include, for example, the information for
generating a scaling list described above, and information
concerning intra prediction and information concerning inter
prediction, which are contained in the block header. The
81
lossless decoding unit 312 outputs the decoded quantized
data and the information for generating a scaling list to
the dequantization/inverse orthogonal transform unit 313.
The lossless decoding unit 312 further outputs the
information concerning intra prediction to the intra
prediction unit 330. The lossless decoding unit 312 further
outputs the information concerning inter prediction to the
motion compensation unit 340.
[0200]
The dequantization/inverse orthogonal transform unit
313 performs dequantization and an inverse orthogonal
transform on the quantized data input from the lossless
decoding unit 312 to generate prediction error data. After
that, the dequantization/inverse orthogonal transform unit
313 outputs the generated prediction error data to the adder
unit 315.
[0201]
The adder unit 315 adds together the prediction error
data input from the dequantization/inverse orthogonal
transform unit 313 and prediction image data input from the
selector 321 to generate decoded image data. After that,
the adder unit 315 outputs the generated decoded image data
to the deblocking filter 316 and the frame memory 319.
[0202]
The deblocking filter 316 filters the decoded image
82
data input from the adder unit 315 to remove blocking
artifacts, and outputs the filtered decoded image data to
the rearrangement buffer 317 and the frame memory 319.
[0203]
The rearrangement buffer 317 rearranges images input
from the deblocking filter 316 to generate a time-series
image data sequence. After that, the rearrangement buffer
317 outputs the generated image data to the D/A conversion
unit 318.
[0204]
The D/A conversion unit 318 converts the image data in
digital form which is input from the rearrangement buffer
317 to an image signal in analog form. After that, the D/A
conversion unit 318 outputs the analog image signal to, for
example, a display (not illustrated) connected to the image
decoding device 300 to display an image.
[0205]
The frame memory 319 stores the decoded image data to
be filtered, which is input from the adder unit 315, and the
filtered decoded image data input from the deblocking filter
316, using a storage medium.
[0206]
The selector 320 switches the destination to which the
image data supplied from the frame memory 319 is to be
output between the intra prediction unit 330 and the motion
83
compensation unit 340, for each block in the image, in
accordance with mode information acquired by the lossless
decoding unit 312. For example, if an intra-prediction mode
is specified, the selector 320 outputs the decoded image
data to be filtered, which is supplied from the frame memory
319, to the intra prediction unit 330 as reference image
data. Furthermore, if an inter-prediction mode is specified,
the selector 320 outputs the filtered decoded image data
supplied from the frame memory 319 to the motion
compensation unit 340 as reference image data.
[0207]
The selector 321 switches the source from which
prediction image data to be supplied to the adder unit 315
is to be output between the intra prediction unit 330 and
the motion compensation unit 340, for each block in the
image, in accordance with mode information acquired by the
lossless decoding unit 312. For example, if the intraprediction
mode is specified, the selector 321 supplies the
prediction image data output from the intra prediction unit
330 to the adder unit 315. If the inter-prediction mode is
specified, the selector 321 supplies the prediction image
data output from the motion compensation unit 340 to the
adder unit 315.
[0208]
The intra prediction unit 330 performs intra-screen
84
prediction of a pixel value based on the information
concerning intra prediction, which is input from the
lossless decoding unit 312, and the reference image data
supplied from the frame memory 319, and generates prediction
image data. After that, the intra prediction unit 330
outputs the generated prediction image data to the selector
321.
[0209]
The motion compensation unit 340 performs a motion
compensation process based on the information concerning
inter prediction, which is input from the lossless decoding
unit 312, and the reference image data supplied from the
frame memory 319, and generates prediction image data.
After that, the motion compensation unit 340 outputs the
generated prediction image data to the selector 321.
[0210]
<2-9. Example configuration of dequantization/inverse
orthogonal transform unit>
Fig. 23 is a block diagram illustrating an example of a
main configuration of the dequantization/inverse orthogonal
transform unit 313 of the image decoding device 300
illustrated in Fig. 22. Referring to Fig. 23, the
dequantization/inverse orthogonal transform unit 313
includes a matrix generation unit 410, a selection unit 430,
a dequantization unit 440, and an inverse orthogonal
85
transform unit 450.
[0211]
(1) Matrix generation unit
The matrix generation unit 410 decodes encoded scaling
list data which is extracted from a bit stream and supplied
by the lossless decoding unit 312, and generates a scaling
list. The matrix generation unit 410 supplies the generated
scaling list to the dequantization unit 440.
[0212]
(2) Selection unit
The selection unit 430 selects a transform unit (TU) to
be used for the inverse orthogonal transform of image data
to be decoded from among a plurality of transform units
having different sizes. Examples of possible sizes of
transform units selectable by the selection unit 430 include
4x4 and 8x8 for H.264/AVC, and include 4x4, 8x8, 16x16, and
32x32 for HEVC. The selection unit 430 may select a
transform unit in accordance with, for example, the LCU, SCU,
and splitflag contained in the header of the encoded stream.
After that, the selection unit 430 outputs information
specifying the size of the selected transform unit to the
dequantization unit 440 and the inverse orthogonal transform
unit 450.
[0213]
(3) Dequantization unit
86
The dequantization unit 440 dequantizes transform
coefficient data quantized when the images are encoded, by
using a scaling list of the transform unit selected by the
selection unit 430. After that, the dequantization unit 440
outputs the dequantized transform coefficient data to the
inverse orthogonal transform unit 450.
[0214]
(4) Inverse orthogonal transform unit
The inverse orthogonal transform unit 450 performs an
inverse orthogonal transform on the transform coefficient
data dequantized by the dequantization unit 440 in units of
the selected transform unit in accordance with the
orthogonal transform scheme used for encoding to generate
prediction error data. After that, the inverse orthogonal
transform unit 450 outputs the generated prediction error
data to the adder unit 315.
[0215]
<2-10. Detailed example configuration of matrix
generation unit>
Fig. 24 is a block diagram illustrating an example of a
detailed configuration of the matrix generation unit 410
illustrated in Fig. 23. Referring to Fig. 24, the matrix
generation unit 410 includes a parameter analysis unit 531,
a prediction unit 532, an entropy decoding unit 533, a
scaling list restoration unit 534, an output unit 535, and a
87
storage unit 536.
[0216]
(1) Parameter analysis unit
The parameter analysis unit 531 analyzes the various
flags and parameters concerning the scaling list, which are
supplied from the lossless decoding unit 312. Furthermore,
in accordance with the analysis results, the parameter
analysis unit 531 supplies various kinds of information
supplied from the lossless decoding unit 312, such as
encoded data of the difference matrix, to the prediction
unit 532 or the entropy decoding unit 533.
[0217]
For example, if pred_mode is equal to 0, the parameter
analysis unit 531 determines that the current mode is the
copy mode, and supplies pred_matrix_id_delta to a copy unit
541. Furthermore, for example, if pred_mode is equal to 1,
the parameter analysis unit 531 determines that the current
mode is a full-scan mode (normal mode), and supplies
pred_matrix_id_delta and pred_size_id_delta to a prediction
matrix generation unit 542.
[0218]
Furthermore, for example, if residual_flag is true, the
parameter analysis unit 531 supplies the encoded data
(exponential Golomb codes) of the scaling list supplied from
the lossless decoding unit 312 to an exp-G unit 551 of the
88
entropy decoding unit 533. The parameter analysis unit 531
further supplies residual_symmetry_flag to the exp-G unit
551.
[0219]
Furthermore, the parameter analysis unit 531 supplies
residual_down_sampling_flag to a difference matrix size
transformation unit 562 of the scaling list restoration unit
534.
[0220]
(2) Prediction unit
The prediction unit 532 generates a prediction matrix
in accordance with the control of the parameter analysis
unit 531. As illustrated in Fig. 24, the prediction unit
532 includes the copy unit 541 and the prediction matrix
generation unit 542.
[0221]
In the copy mode, the copy unit 541 copies a previously
transmitted scaling list, and uses the copied scaling list
as a prediction matrix. More specifically, the copy unit
541 reads a previously transmitted scaling list
corresponding to pred_matrix_id_delta and having the same
size as the scaling list for the current region from the
storage unit 536, uses the read scaling list as a prediction
image, and supplies the prediction image to the output unit
535.
89
[0222]
In the normal mode, the prediction matrix generation
unit 542 generates (or predicts) a prediction matrix using a
previously transmitted scaling list. More specifically, the
prediction matrix generation unit 542 reads a previously
transmitted scaling list corresponding to
pred_matrix_id_delta and pred_size_id_delta from the storage
unit 536, and generates a prediction matrix using the read
scaling list. In other words, the prediction matrix
generation unit 542 generates a prediction matrix similar to
the prediction matrix generated by the prediction matrix
generation unit 172 (Fig. 16) of the image encoding device
10. The prediction matrix generation unit 542 supplies the
generated prediction matrix to a prediction matrix size
transformation unit 561 of the scaling list restoration unit
534.
[0223]
(3) Entropy decoding unit
The entropy decoding unit 533 restores a difference
matrix from the exponential Golomb codes supplied from the
parameter analysis unit 531. As illustrated in Fig. 24, the
entropy decoding unit 533 includes the exp-G unit 551, an
inverse DPCM unit 552, and an inverse overlap determination
unit 553.
[0224]
90
The exp-G unit 551 decodes the signed or unsigned
exponential Golomb codes (hereinafter also referred to as
exponential Golomb decoding) to restore DPCM data. The exp-
G unit 551 supplies the restored DPCM data together with
residual_symmetry_flag to the inverse DPCM unit 552.
[0225]
The inverse DPCM unit 552 performs DPCM decoding of
data from which the overlapping portion has been removed to
generate residual data from the DPCM data. The inverse DPCM
unit 552 supplies the generated residual data together with
residual_symmetry_flag to the inverse overlap determination
unit 553.
[0226]
If residual_symmetry_flag is true, that is, if the
residual data is a remaining portion of a 135-degree
symmetric matrix from which the data (matrix elements) of
the overlapping symmetric part has been removed, the inverse
overlap determination unit 553 restores the data of the
symmetric part. In other words, a difference matrix of a
135-degree symmetric matrix is restored. Note that if
residual_symmetry_flag is not true, that is, if the residual
data represents a matrix that is not a 135-degree symmetric
matrix, the inverse overlap determination unit 553 uses the
residual data as a difference matrix without restoring data
of a symmetric part. The inverse overlap determination unit
91
553 supplies the difference matrix restored in the way
described above to the scaling list restoration unit 534
(the difference matrix size transformation unit 562).
[0227]
(4) Scaling list restoration unit
The scaling list restoration unit 534 restores a
scaling list. As illustrated in Fig. 24, the scaling list
restoration unit 534 includes the prediction matrix size
transformation unit 561, the difference matrix size
transformation unit 562, a dequantization unit 563, and a
computation unit 564.
[0228]
If the size of the prediction matrix supplied from the
prediction unit 532 (the prediction matrix generation unit
542) is different from the size of the scaling list for the
current region to be restored, the prediction matrix size
transformation unit 561 converts the size of the prediction
matrix.
[0229]
For example, if the size of the prediction matrix is
larger than the size of the scaling list, the prediction
matrix size transformation unit 561 down-converts the
prediction matrix. Furthermore, for example, if the size of
the prediction matrix is smaller than the size of the
scaling list, the prediction matrix size transformation unit
92
561 up-converts the prediction matrix. The same method as
that for the prediction matrix size transformation unit 181
(Fig. 16) of the image encoding device 10 is selected as a
conversion method.
[0230]
The prediction matrix size transformation unit 561
supplies the prediction matrix whose size has been made to
match that of the scaling list to the computation unit 564.
[0231]
If residual_down_sampling_flag is true, that is, if the
size of the transmitted difference matrix is smaller than
the size of the current region to be dequantized, the
difference matrix size transformation unit 562 up-converts
the difference matrix to increase the size of the difference
matrix to a size corresponding to the current region to be
dequantized. Any method for up-conversion may be used. For
example, a method corresponding to the down-conversion
method performed by the difference matrix size
transformation unit 163 (Fig. 16) of the image encoding
device 10 may be used.
[0232]
For example, if the difference matrix size
transformation unit 163 has downsampled the difference
matrix, the difference matrix size transformation unit 562
may upsample the difference matrix. Alternatively, if the
93
difference matrix size transformation unit 163 has subsampled
the difference matrix, the difference matrix size
transformation unit 562 may perform inverse subsampling of
the difference matrix.
[0233]
For example, the difference matrix size transformation
unit 562 may perform a nearest neighbor interpolation
process (nearest neighbor) as illustrated in Fig. 25 rather
than general linear interpolation. The nearest neighbor
interpolation process can reduce memory capacity.
[0234]
Accordingly, even if a scaling list having a large size
is not transmitted, data obtained after upsampling need not
be stored for upsampling from a scaling list having a small
size. In addition, an intermediate buffer or the like is
not necessary when data involved in computation during
upsampling is stored.
[0235]
Note that if residual_down_sampling_flag is not true,
that is, if the difference matrix is transmitted with the
same size as that when used for the quantization process,
the difference matrix size transformation unit 562 omits the
up-conversion of the difference matrix (or may up-convert
the difference matrix by a factor of 1 ) .
[0236]
94
The difference matrix size transformation unit 562
supplies the difference matrix up-converted in the manner
described above, as necessary, to the dequantization unit
563.
[0237]
The dequantization unit 563 dequantizes the supplied
difference matrix (quantized data) using a method
corresponding to that for quantization performed by the
quantization unit 183 (Fig. 16) of the image encoding device
10, and supplies the dequantized difference matrix to the
computation unit 564. Note that if the quantization unit
183 is omitted, that is, if the difference matrix supplied
from the difference matrix size transformation unit 562 is
not quantized data, the dequantization unit 563 can be
omitted.
[0238]
The computation unit 564 adds together the prediction
matrix supplied from the prediction matrix size
transformation unit 561 and the difference matrix supplied
from the dequantization unit 563, and restores a scaling
list for the current region. The computation unit 564
supplies the restored scaling list to the output unit 535
and the storage unit 536.
[0239]
(5) Output unit
95
The output unit 535 outputs the supplied information to
a device outside the matrix generation unit 410. For
example, in the copy mode, the output unit 535 supplies the
prediction matrix supplied from the copy unit 541 to the
dequantization unit 440 as a scaling list for the current
region. Furthermore, for example, in the normal mode, the
output unit 535 supplies the scaling list for the current
region supplied from the scaling list restoration unit 534
(the computation unit 564) to the dequantization unit 440.
[0240]
(6) Storage unit
The storage unit 536 stores the scaling list supplied
from the scaling list restoration unit 534 (the computation
unit 564) together with the size and the list ID of the
scaling list. The information concerning the scaling list
stored in the storage unit 536 is used to generate
prediction matrices of other orthogonal transform units
which are processed later in time. In other words, the
storage unit 536 supplies the stored information concerning
the scaling list to the prediction unit 532 as information
concerning a previously transmitted scaling list.
[0241]
<2-11. Detailed example configuration of inverse DPCM
unit>
Fig. 26 is a block diagram illustrating an example of a
96
detailed configuration of the inverse DPCM unit 552
illustrated in Fig. 24. Referring to Fig. 26, the inverse
DPCM unit 552 includes an initial setting unit 571, a DPCM
decoding unit 572, and a DC coefficient extraction unit 573.
[0242]
The initial setting unit 571 acquires sizeID and
MatrixID, and sets various variables to initial values. The
initial setting unit 571 supplies the acquired and set
information to the DPCM decoding unit 572.
[0243]
The DPCM decoding unit 572 determines individual
coefficients (the DC coefficient and the AC coefficients)
from the difference values (scaling_list_delta_coef) of the
DC coefficient and the AC coefficients using the initial
settings and the like supplied from the initial setting unit
571. The DPCM decoding unit 572 supplies the determined
coefficients to the DC coefficient extraction unit 573
(ScalingList[i]).
[0244]
The DC coefficient extraction unit 573 extracts the DC
coefficient from among the coefficients (ScalingList[i])
supplied from the DPCM decoding unit 572. The DC
coefficient is located at the beginning of the AC
coefficients. That is, the initial coefficient
(ScalingList[0]) among the coefficients supplied from the
97
DPCM decoding unit 572 is the DC coefficient. The DC
coefficient extraction unit 573 extracts the coefficient
located at the beginning as the DC coefficient, and outputs
the extracted coefficient to the inverse overlap
determination unit 553 (DC_coef). The DC coefficient
extraction unit 573 outputs the other coefficients
(ScalingList[i] (i > 0)) to the inverse overlap
determination unit 553 as the AC coefficients.
[0245]
Accordingly, the inverse DPCM unit 552 can perform
correct DPCM decoding, and can obtain the DC coefficient and
the AC coefficients. That is, the image decoding device 300
can suppress an increase in the amount of coding of a
scaling list.
[0246]
<2-12. Flow of quantization matrix decoding process>
An example of the flow of a quantization matrix
decoding process executed by the matrix generation unit 410
having the configuration described above will be described
with reference to a flowchart illustrated in Fig. 27.
[0247]
When the quantization matrix decoding process is
started, in step S301, the parameter analysis unit 531 reads
the quantized values (Qscale0 to Qscale3) of regions 0 to 3.
[0248]
98
In step S302, the parameter analysis unit 531 reads
pred_mode. In step S303, the parameter analysis unit 531
determines whether or not pred_mode is equal to 0. If it is
determined that pred_mode is equal to 0, the parameter
analysis unit 531 determines that the current mode is the
copy mode, and advances the process to step S304.
[0249]
In step S304, the parameter analysis unit 531 reads
pred_matrix_id_delta. In step S305, the copy unit 541
copies a scaling list that has been transmitted, and uses
the copied scaling list as a prediction matrix. In the copy
mode, the prediction matrix is output as the scaling list
for the current region. When the processing of step S305 is
completed, the copy unit 541 ends the quantization matrix
decoding process.
[0250]
Furthermore, if it is determined in step S303 that
pred_mode is not equal to 0, the parameter analysis unit 531
determines that the current mode is the full-scan mode
(normal mode), and advances the process to step S306.
[0251]
In step S306, the parameter analysis unit 531 reads
pred_matrix_id_delta, pred_size_id_delta, and residual_flag.
In step S307, the prediction matrix generation unit 542
generates a prediction matrix from a scaling list that has
99
been transmitted.
[0252]
In step S308, the parameter analysis unit 531
determines whether or not residual_flag is true. If it is
determined that residual_flag is not true, no residual
matrices exist, and the prediction matrix generated in step
S307 is output as the scaling list for the current region.
In this case, therefore, the parameter analysis unit 531
ends the quantization matrix decoding process.
[0253]
Furthermore, if it is determined in step S308 that
residual_flag is true, the parameter analysis unit 531
advances the process to step S309.
[0254]
In step S309, the parameter analysis unit 531 reads
residual_down_sampling_flag and residual_symmetry_flag.
[0255]
In step S310, the exp-G unit 551 and the inverse DPCM
unit 552 decode the exponential Golomb codes of the residual
matrix, and generate residual data.
[0256]
In step S311, the inverse overlap determination unit
553 determines whether or not residual_symmetry_flag is true.
If it is determined that residual_symmetry_flag is true, the
inverse overlap determination unit 553 advances the process
100
to step S312, and restores the removed overlapping portion
of the residual data (or performs an inverse symmetry
process). When a difference matrix that is a 135-degree
symmetric matrix is generated in the way described above,
the inverse overlap determination unit 553 advances the
process to step S313.
[0257]
Furthermore, if it is determined in step S311 that
residual_symmetry_flag is not true (or if the residual data
is a difference matrix that is not a 135-degree symmetric
matrix), the inverse overlap determination unit 553 advances
the process to step S313 while skipping the processing of
step S312 (or without performing an inverse symmetry
process).
[0258]
In step S313, the difference matrix size transformation
unit 562 determines whether or not
residual_down_sampling_flag is true. If it is determined
that residual_down_sampling_flag is true, the difference
matrix size transformation unit 562 advances the process to
step S314, and up-converts the difference matrix to a size
corresponding to the current region to be dequantized.
After the difference matrix is up-converted, the difference
matrix size transformation unit 562 advances the process to
step S315.
101
[0259]
Furthermore, if it is determined in step S313 that
residual_down_sampling_flag is not true, the difference
matrix size transformation unit 562 advances the process to
step S315 while skipping the processing of step S314 (or
without up-converting the difference matrix).
[0260]
In step S315, the computation unit 564 adds the
difference matrix to the prediction matrix to generate a
scaling list for the current region. When the processing of
step S315 is completed, the quantization matrix decoding
process ends.
[0261]
<2-13. Flow of residual signal decoding process>
Next, an example of the flow of the residual signal
decoding process executed in step S310 in Fig. 27 will be
described with reference to a flowchart illustrated in Fig.
28.
[0262]
When the residual signal decoding process is started,
in step S331, the exp-G unit 551 decodes the supplied
exponential Golomb codes.
[0263]
In step S332, the inverse DPCM unit 552 performs an
inverse DPCM process on DPCM data obtained by the exp-G unit
102
551 through decoding.
[0264]
When the inverse DPCM process is completed, the inverse
DPCM unit 552 ends the residual signal decoding process, and
returns the process to Fig. 27.
[0265]
<2-14. Flow of inverse DPCM process>
Next, an example of the flow of the inverse DPCM
process executed in step S332 in Fig. 28 will be described
with reference to a flowchart illustrated in Fig. 29.
[0266]
When the inverse DPCM process is started, in step S351,
the initial setting unit 571 acquires sizeID and MatrixID.
[0267]
In step S352, the initial setting unit 571 sets coefNum
as follows.
coefNum = min((1<<(4+(sizeID<<1))), 65)
[0268]
In step S353, the initial setting unit 571 sets a
variable i and a variable nextcoef as follows.
i = 0
nextcoef = 8
[0269]
In step S354, the DPCM decoding unit 572 determines
whether or not variable i < coefNum. If the variable i is
103
smaller than coefNum, the initial setting unit 571 advances
the process to step S355.
[0270]
In step S355, the DPCM decoding unit 572 reads DPCM
data of the coefficient (scaling_list_delta_coef).
[0271]
In step S356, the DPCM decoding unit 572 determines
nextcoef as below using the read DPCM data, and further
determines scalingList[i].
nextcoef = (nextcoef + scaling_list_delta_coef + 256) %
256
scalingList[i] = nextcoef
[0272]
In step S357, the DC coefficient extraction unit 573
determines whether or not sizeID is larger than 1 and
whether or not the variable i is equal to 0 (that is, the
coefficient located at the beginning). If it is determined
that sizeID is larger than 1 and the variable i represents
the coefficient located at the beginning, the DC coefficient
extraction unit 573 advances the process to step S358, and
uses the coefficient as the DC coefficient (DC_coef =
nextcoef). When the processing of step S358 is completed,
the DC coefficient extraction unit 573 advances the process
to step S360.
[0273]
104
Furthermore, if it is determined in step S357 that
sizeID is less than or equal to 1 or that the variable i
does not represent the coefficient located at the beginning,
the DC coefficient extraction unit 573 advances the process
to step S359, and shifts the variable i for each coefficient
by one because the DC coefficient has been extracted.
(ScalingList[(i-(sizeID)>1)?1;0] = nextcoef) If the
processing of step S359 is completed, the DC coefficient
extraction unit 573 advances the process to step S360.
[0274]
In step S360, the DPCM decoding unit 572 increments the
variable i to change the processing target to the subsequent
coefficient, and then returns the process to step S354.
[0275]
In step S354, the processing of steps S354 to S360 is
repeatedly performed until it is determined that the
variable i is greater than or equal to coefNum. If it is
determined in step S354 that the variable i is greater than
or equal to coefNum, the DPCM decoding unit 572 ends the
inverse DPCM process, and returns the process to Fig. 28.
[0276]
Accordingly, the difference between the DC coefficient
and the AC coefficient located at the beginning of the AC
coefficients may be correctly decoded. Therefore, the image
decoding device 300 can suppress an increase in the amount
105
of coding of a scaling list.
[0277]
<3. Third Embodiment>
<3-1. Syntax: Second method>
Another method for transmitting a difference between
the DC coefficient and another coefficient, instead of the
DC coefficient, may be to, for example, transmit a
difference between the DC coefficient and the (0, 0)
component of an 8x8 matrix as DPCM data different from the
DPCM data of the 8x8 matrix (second method). For example,
after DPCM transmission of an 8x8 matrix, the difference
between the DC coefficient and the (0, 0) component of the
8x8 matrix may be transmitted.
[0278]
Accordingly, similarly to the first method, the
compression ratio can be more improved when the value of the
(0, 0) coefficient (AC coefficient) of an 8x8 matrix and the
value of the DC coefficient are close to each other.
[0279]
Fig. 30 illustrates the syntax of a scaling list in the
second method. In the example illustrated in Fig. 30, 64
difference values (scalinglistdeltacoef) between
coefficients are read. Finally, the difference
(scalinglistdccoefdelta) between the DC coefficient and
the (0, 0) coefficient (AC coefficient) is read, and the DC
106
coefficient is determined from the difference.
[0280]
In the second method, accordingly, syntax for decoding
AC coefficients can be similar to that of the related art
illustrated in Fig. 12. That is, the syntax for the second
method can be obtained by modifying the example of the
related art by a small amount, and can be more feasible than
that for the first method.
[0281]
However, whereas the second method does not allow an
image decoding device to obtain the DC coefficient until the
image decoding device have received all the coefficients and
have decompressed all the DPCM data, the first method allows
an image decoding device to restore the DC coefficient at
the time when the image decoding device receives the initial
coefficient.
[0282]
An image encoding device that implements the syntax for
the second method described above will be described
hereinafter.
[0283]
<3-2. Detailed example configuration of DPCM unit>
In the second method, the image encoding device 10 has
a configuration basically similar to that in the first
method described above. Specifically, the image encoding
107
device 10 has a configuration as in the example illustrated
in Fig. 14. Further, the orthogonal transform/quantization
unit 14 has a configuration as in the example illustrated in
Fig. 15. Further, the matrix processing unit 150 has a
configuration as in the example illustrated in Fig. 16.
[0284]
An example configuration of the DPCM unit 192 in the
second example is illustrated in Fig. 31. As illustrated in
Fig. 31, in the second example, the DPCM unit 192 includes
an AC coefficient buffer 611, an AC coefficient encoding
unit 612, an AC coefficient DPCM unit 613, and a DC
coefficient DPCM unit 614.
[0285]
The AC coefficient buffer 611 stores the initial AC
coefficient (that is, the (0, 0) coefficient) supplied from
the overlap determination unit 191. The AC coefficient
buffer 611 supplies the stored initial AC coefficient (AC
coefficient (0, 0)) to the DC coefficient DPCM unit 614 at a
predetermined timing after all the AC coefficients have been
subjected to a DPCM process, or in response to a request.
[0286]
The AC coefficient encoding unit 612 acquires the
initial AC coefficient (AC coefficient (0, 0)) supplied from
the overlap determination unit 191, and subtracts the value
of the initial AC coefficient from a constant (for example,
108
8 ) . The AC coefficient encoding unit 612 supplies a
subtraction result (difference) to the exp-G unit 193 as the
initial coefficient (scaling_list_delta_coef (i = 0)) of the
DPCM data of the AC coefficients.
[0287]
The AC coefficient DPCM unit 613 acquires the AC
coefficients supplied from the overlap determination unit
191, determines, for each of the second and subsequent AC
coefficients, the difference (DPCM) from the immediately
preceding AC coefficient, and supplies the determined
differences to the exp-G unit 193 as DPCM data
(scaling_list_delta_coef (i = 1 to 63)).
[0288]
The DC coefficient DPCM unit 614 acquires the DC
coefficient supplied from the overlap determination unit 191.
The DC coefficient DPCM unit 614 further acquires the
initial AC coefficient (AC coefficient (0, 0)) held in the
AC coefficient buffer 611. The DC coefficient DPCM unit 614
subtracts the initial AC coefficient (AC coefficient (0, 0))
from the DC coefficient to determine the difference
therebetween, and supplies the determined difference to the
exp-G unit 193 as DPCM data of the DC coefficient
(scaling_list_dc_coef_delta).
[0289]
As described above, in the second method, a difference
109
between the DC coefficient and another coefficient (the
initial AC coefficient) is determined. Then, the difference
is transmitted, as DPCM data of the DC coefficient
(scaling_list_dc_coef_delta) different from DPCM data of the
AC coefficients, after the transmission of DPCM data of the
AC coefficients (scaling_list_delta_coef) that is a
difference between the AC coefficients. Accordingly,
similarly to the first method, the image encoding device 10
can improve the coding efficiency of a scaling list.
[0290]
<3-3. Flow of DPCM process>
Also in the second method, the image encoding device 10
executes a quantization matrix encoding process in a manner
similar to that in the first method described with reference
to the flowchart illustrated in Fig. 20.
[0291]
An example of the flow of a DPCM process in the second
method, which is executed in step S112 in Fig. 20, will be
described with reference to a flowchart illustrated in Fig.
32.
[0292]
When the DPCM process is started, in step S401, the AC
coefficient buffer 611 holds the initial AC coefficient.
[0293]
In step S402, the AC coefficient encoding unit 612
110
subtracts the initial AC coefficient from a predetermined
constant (for example, 8) to determine the difference
therebetween (initial DPCM data).
[0294]
The processing of steps S403 to S405 is executed by the
AC coefficient DPCM unit 613 in a manner similar to the
processing of steps S133 to S135 in Fig. 21. That is, the
processing of steps S403 to S405 is repeatedly executed to
generate DPCM data of all the AC coefficients (the
differences from the immediately preceding AC coefficients).
[0295]
If it is determined in step S403 that all the AC
coefficients have been processed (that is, if there is no
unprocessed AC coefficient), the AC coefficient DPCM unit
613 advances the process to step S406.
[0296]
In step S406, the DC coefficient DPCM unit 614
subtracts the initial AC coefficient held in step S401 from
the DC coefficient to determine a difference therebetween
(DPCM data for the DC coefficient).
[0297]
When the processing of step S406 is completed, the DC
coefficient DPCM unit 614 ends the DPCM process, and returns
the process to Fig. 20.
[0298]
111
Accordingly, a difference between the DC coefficient
and another coefficient is also determined and transmitted
to an image decoding device as DPCM data. Thus, the image
encoding device 10 can suppress an increase in the amount of
coding of a scaling list.
[0299]
<3-4. Detailed example configuration of inverse DPCM
unit>
In the second method, the image decoding device 300 has
a configuration basically similar to that in the first
method. Specifically, also in the second method, the image
decoding device 300 has a configuration as in the example
illustrated in Fig. 22. Furthermore, the
dequantization/inverse orthogonal transform unit 313 has a
configuration as in the example illustrated in Fig. 23.
Moreover, the matrix generation unit 410 has a configuration
as in the example illustrated in Fig. 24.
[0300]
Fig. 33 is a block diagram illustrating an example of a
detailed configuration of the inverse DPCM unit 552
illustrated in Fig. 24 in the second method. Referring to
Fig. 33, the inverse DPCM unit 552 includes an initial
setting unit 621, an AC coefficient DPCM decoding unit 622,
an AC coefficient buffer 623, and a DC coefficient DPCM
decoding unit 624.
112
[0301]
The initial setting unit 621 acquires sizeID and
MatrixID, and sets various variables to initial values. The
initial setting unit 621 supplies the acquired and set
information to the AC coefficient DPCM decoding unit 622.
[0302]
The AC coefficient DPCM decoding unit 622 acquires the
DPCM data of the AC coefficients (scaling_list_delta_coef)
supplied from the exp-G unit 551. The AC coefficient DPCM
decoding unit 622 decodes the acquired DPCM data of the AC
coefficients using the initial settings and the like
supplied from the initial setting unit 621 to determine AC
coefficients. The AC coefficient DPCM decoding unit 622
supplies the determined AC coefficients (ScalingList[i]) to
the inverse overlap determination unit 553. The AC
coefficient DPCM decoding unit 622 further supplies the
initial AC coefficient (ScalingList[0], that is, the AC
coefficient (0, 0)) among the determined AC coefficients to
the AC coefficient buffer 623 for holding.
[0303]
The AC coefficient buffer 623 stores the initial AC
coefficient (ScalingList[0], that is, the AC coefficient (0,
0)) supplied from the AC coefficient DPCM decoding unit 622.
The AC coefficient buffer 623 supplies the initial AC
coefficient (ScalingList[0], that is, the AC coefficient (0,
113
0)) to the DC coefficient DPCM decoding unit 624 at a
predetermined timing or in response to a request.
[0304]
The DC coefficient DPCM decoding unit 624 acquires the
DPCM data of the DC coefficient (scaling_list_dc_coef_delta)
supplied from the exp-G unit 551. The DC coefficient DPCM
decoding unit 624 further acquires the initial AC
coefficient (ScalingList[0], that is, the AC coefficient (0,
0)) stored in the AC coefficient buffer 623. The DC
coefficient DPCM decoding unit 624 decodes the DPCM data of
the DC coefficient using the initial AC coefficient to
determine the DC coefficient. The DC coefficient DPCM
decoding unit 624 supplies the determined DC coefficient
(DC_coef) to the inverse overlap determination unit 553.
[0305]
Accordingly, the inverse DPCM unit 552 can perform
correct DPCM decoding, and can obtain the DC coefficient and
the AC coefficients. That is, the image decoding device 300
can suppress an increase in the amount of coding of a
scaling list.
[0306]
<3-5. Flow of inverse DPCM process>
Also in the second method, the image decoding device
300 executes a quantization matrix decoding process in a
manner similar to that in the first method described above
114
with reference to the flowchart illustrated in Fig. 27.
Similarly, the image decoding device 300 executes a residual
signal decoding process in a manner similar to that in the
first method described above with reference to the flowchart
illustrated in Fig. 28.
[0307]
An example of the flow of the inverse DPCM process
executed by the inverse DPCM unit 552 will be described with
reference to a flowchart illustrated in Fig. 34.
[0308]
When the inverse DPCM process is started, in step S421,
the initial setting unit 621 acquires sizeID and MatrixID.
[0309]
In step S422, the initial setting unit 621 sets coefNum
as follows.
coefNum = min((1<<(4+(sizeID<<1))), 64)
[0310]
In step S423, the initial setting unit 621 sets a
variable i and a variable nextcoef as follows.
i = 0
nextcoef = 8
[0311]
In step S424, the DPCM decoding unit 572 determines
whether or not variable i < coefNum. If the variable i is
smaller than coefNum, the initial setting unit 621 advances
115
the process to step S425.
[0312]
In step S425, the AC coefficient DPCM decoding unit 622
reads DPCM data of the AC coefficients
(scaling_list_delta_coef).
[0313]
In step S426, the AC coefficient DPCM decoding unit 622
determines nextcoef as below using the read DPCM data, and
further determines scalingList[i].
nextcoef = (nextcoef + scaling_list_delta_coef + 256) %
256
scalingList[i] = nextcoef
Note that the calculated initial AC coefficient
(ScalingList[0], that is, the AC coefficient (0, 0)) is held
in the AC coefficient buffer 623.
[0314]
In step S427, the AC coefficient DPCM decoding unit 622
increments the variable i to change the target to be
processed to the subsequent coefficient, and then returns
the process to step S424.
[0315]
In step S424, the processing of steps S424 to S427 is
repeatedly performed until it is determined that the
variable i is greater than or equal to coefNum. If it is
determined in step S424 that the variable i is greater than
116
or equal to coefNum, the AC coefficient DPCM decoding unit
622 advances the process to step S428.
[0316]
In step S428, the DC coefficient DPCM decoding unit 624
determines whether or not sizeID is greater than 1. If it
is determined that sizeID is greater than 1, the DC
coefficient DPCM decoding unit 624 advances the process to
step S429, and reads the DPCM data of the DC coefficient
(scaling_list_dc_coef_delta).
[0317]
In step S430, the DC coefficient DPCM decoding unit 624
acquires the initial AC coefficient (ScalingList[0], that is,
the AC coefficient (0, 0)) held in the AC coefficient buffer
623, and decodes the DPCM data of the DC coefficient
(DC_coef) using the initial AC coefficient as follows.
DC_coef = scaling_list_dc_coef_delta + ScalingList[0]
[0318]
When the DC coefficient (DC_coef) is obtained, the DC
coefficient DPCM decoding unit 624 ends the inverse DPCM
process, and returns the process to Fig. 28.
[0319]
Furthermore, if it is determined in step S428 that
sizeID is less than or equal to 1, the DC coefficient DPCM
decoding unit 624 ends the inverse DPCM process, and returns
the process to Fig. 28.
117
[0320]
Accordingly, the difference between the DC coefficient
and the AC coefficient located at the beginning of the AC
coefficients can be correctly decoded. Therefore, the image
decoding device 300 can suppress an increase in the amount
of coding of a scaling list.
[0321]
<4. Fourth Embodiment>
<4-1. Syntax: Third method>
In the second method described above, the DC
coefficient may also be limited to a value smaller than the
initial AC coefficient (AC coefficient (0, 0)) (third
method).
[0322]
This ensures that the DPCM data of the DC coefficient,
that is, a difference value obtained by subtracting the
initial AC coefficient from the DC coefficient, can be a
positive value. This DPCM data can thus be encoded using
unsigned exponential Golomb codes. Therefore, the third
method may prevent the DC coefficient from being larger than
the initial AC coefficient, but can reduce the amount of
coding compared to the first method and the second method.
[0323]
Fig. 35 illustrates the syntax of a scaling list in the
third method. As illustrated in Fig. 35, in this case, the
118
DPCM data of the DC coefficient (scaling_list_dc_coef_delta)
is limited to a positive value.
[0324]
The syntax for the third method described above can be
implemented by an image encoding device 10 similar to that
in the second method. In the third method, however, the
exp-G unit 193 can encode the DPCM data of the DC
coefficient using unsigned exponential Golomb codes. Note
that the image encoding device 10 can execute processes such
as a quantization matrix encoding process and a DPCM process
in a manner similar to that in the second method.
[0325]
Furthermore, the syntax for the third method can be
implemented by the image decoding device 300 in a manner
similar to that in the second method. Moreover, the image
decoding device 300 can execute a quantization matrix
decoding process in a manner similar to that in the second
method.
[0326]
<4-2. Flow of inverse DPCM process>
An example of the flow of an inverse DPCM process
executed by the inverse DPCM unit 552 will be described with
reference to a flowchart illustrated in Fig. 36.
[0327]
The processing of steps S451 to S459 is performed in a
119
manner similar to the processing of steps S421 to S429 in
Fig. 34.
[0328]
In step S460, the DC coefficient DPCM decoding unit 624
acquires the initial AC coefficient (ScalingList[0], that is,
the AC coefficient (0, 0)) held in the AC coefficient buffer
623, and decodes the DPCM data of the DC coefficient
(DC_coef) as below using the initial AC coefficient.
DC_coef = ScalingList[0] - scaling_list_dc_coef_delta
[0329]
When the DC coefficient (DC_coef) is obtained, the DC
coefficient DPCM decoding unit 624 ends the inverse DPCM
process, and returns the process to Fig. 28.
[0330]
Furthermore, if it is determined in step S458 that
sizeID is less than or equal to 1, the DC coefficient DPCM
decoding unit 624 ends the inverse DPCM process, and returns
the process to Fig. 28.
[0331]
Accordingly, the difference between the DC coefficient
and the AC coefficient located at the beginning of the AC
coefficients can be correctly decoded. Therefore, the image
decoding device 300 can suppress an increase in the amount
of coding of a scaling list.
[0332]
120
<5. Fifth Embodiment>
<5-1. Syntax: Fourth method>
Another method for transmitting a difference between
the DC coefficient and another coefficient, instead of the
DC coefficient, may be to, for example, collect only the DC
coefficients of a plurality of scaling lists and to perform
DPCM by taking differences between the DC coefficients
separately from the AC coefficients of the individual
scaling lists (fourth method). In this case, DPCM data of
the DC coefficients is a collection of pieces of data for
the plurality of scaling lists, and is transmitted as data
different from DPCM data of the AC coefficients of the
individual scaling lists.
[0333]
Accordingly, the compression ratio can be more improved
when, for example, there are correlations between the DC
coefficients of the scaling lists (MatrixID).
[0334]
Fig. 37 illustrates the syntax for the DC coefficient
of a scaling list in the fourth method. In this case, since
the DC coefficients are processed in cycles different from
those for the AC coefficients of the individual scaling
lists, as illustrated in the example illustrated in Fig. 37,
processes for the AC coefficients and processes for the DC
coefficients need to be independent from each other.
121
[0335]
This ensures that more various methods for scaling list
encoding and decoding processes can be achieved although the
complexity of the DPCM process and the inverse DPCM process
may be increased. For example, a process for copying only
the AC coefficients and making the values of the DC
coefficients different in the copy mode can be easily
implemented.
[0336]
The number of scaling lists in which the DC
coefficients are collectively processed is arbitrary.
[0337]
<5-2. Detailed example configuration of DPCM unit>
In the fourth method, the image encoding device 10 has
a configuration basically similar to that in the first
method described above. Specifically, the image encoding
device 10 has a configuration as in the example illustrated
in Fig. 14. Furthermore, the orthogonal
transform/quantization unit 14 has a configuration as in the
example illustrated in Fig. 15. Moreover, the matrix
processing unit 150 has a configuration as in the example
illustrated in Fig. 16.
[0338]
An example configuration of the DPCM unit 192 in the
fourth method is illustrated in Fig. 38. As illustrated in
122
Fig. 38, in this case, the DPCM unit 192 includes an AC
coefficient DPCM unit 631, a DC coefficient buffer 632, and
a DC coefficient DPCM unit 633.
[0339]
The AC coefficient DPCM unit 631 performs a DPCM
process of the individual AC coefficients of each scaling
list which are supplied from the overlap determination unit
191. Specifically, the AC coefficient DPCM unit 631
subtracts, for each scaling list, the initial AC coefficient
from a predetermined constant (for example, 8 ) , and
subtracts the AC coefficient being processed (current AC
coefficient) from the immediately preceding AC coefficient.
The AC coefficient DPCM unit 631 supplies DPCM data
(scaling_list_delta_coef) generated for each scaling list to
the exp-G unit 193.
[0340]
The DC coefficient buffer 632 stores the DC
coefficients of the individual scaling lists supplied from
the overlap determination unit 191. The DC coefficient
buffer 632 supplies the stored DC coefficients to the DC
coefficient DPCM unit 633 at a predetermined timing or in
response to a request.
[0341]
The DC coefficient DPCM unit 633 acquires the DC
coefficients accumulated in the DC coefficient buffer 632.
123
The DC coefficient DPCM unit 633 determines DPCM data of the
acquired DC coefficients. Specifically, the DC coefficient
DPCM unit 633 subtracts the initial DC coefficient from a
predetermined constant (for example, 8 ) , and subtracts the
DC coefficient being processed (current DC coefficient) from
the immediately preceding DC coefficient. The DC
coefficient DPCM unit 633 supplies the generated DPCM data
(scaling_list_delta_coef) to the exp-G unit 193.
[0342]
Accordingly, the image encoding device 10 can improve
the coding efficiency of a scaling list.
[0343]
<5-3. Flow of DPCM process>
Also in the fourth method, the image encoding device 10
executes a quantization matrix encoding process in a manner
similar to that in the first method described above with
reference to the flowchart illustrated in Fig. 20.
[0344]
An example of the flow of a DPCM process in the fourth
method, which is executed in step S112 in Fig. 20, will be
described with reference to a flowchart illustrated in Fig.
39.
[0345]
The processing of steps S481 to S485 is executed by the
AC coefficient DPCM unit 631 in a manner similar to the
124
processing of steps S401 to S405 (the processing in the
second method) in Fig. 32.
[0346]
If it is determined in step S483 that all the AC
coefficients have been processed, the AC coefficient DPCM
unit 631 advances the process to step S486.
[0347]
In step S486, the AC coefficient DPCM unit 631
determines whether or not all the scaling lists (or
difference matrices) in which the DC coefficients are
collectively DPCM encoded have been processed. If it is
determined that there is an unprocessed scaling list (or
difference matrix), the AC coefficient DPCM unit 631 returns
the process to step S481.
[0348]
If it is determined in step S486 that all the scaling
lists (or difference matrices) have been processed, the AC
coefficient DPCM unit 631 advances the process to step S487.
[0349]
The DC coefficient DPCM unit 633 executes the
processing of steps S487 to S491 on the DC coefficients
stored in the DC coefficient buffer 632 in a manner similar
to the processing of steps S481 to S485.
[0350]
If it is determined in step S489 that all the DC
125
coefficients stored in the DC coefficient buffer 632 have
been processed, the DC coefficient DPCM unit 633 ends the
DPCM process, and returns the process to Fig. 20.
[0351]
By executing a DPCM process in the manner described
above, the image encoding device 10 can improve the coding
efficiency of a scaling list.
[0352]
<5-4. Detailed example configuration of inverse DPCM
unit>
The image decoding device 300 in the fourth method has
a configuration basically similar to that in the first
method. Specifically, also in the fourth method, the image
decoding device 300 has a configuration as in the example
illustrated in Fig. 22. Further, the dequantization/inverse
orthogonal transform unit 313 has a configuration as in the
example illustrated in Fig. 23. Moreover, the matrix
generation unit 410 has a configuration as in the example
illustrated in Fig. 24.
[0353]
Fig. 40 is a block diagram illustrating an example of a
detailed configuration of the inverse DPCM unit 552
illustrated in Fig. 24 in the fourth method. Referring to
Fig. 40, the inverse DPCM unit 552 includes an initial
setting unit 641, an AC coefficient DPCM decoding unit 642,
126
and a DC coefficient DPCM decoding unit 643.
[0354]
The initial setting unit 641 acquires sizeID and
MatrixID, and sets various variables to initial values. The
initial setting unit 641 supplies the acquired and set
information to the AC coefficient DPCM decoding unit 642 and
the DC coefficient DPCM decoding unit 643.
[0355]
The AC coefficient DPCM decoding unit 642 acquires the
DPCM data of the AC coefficients
(scaling_list_delta_coef(ac)) supplied from the exp-G unit
551. The AC coefficient DPCM decoding unit 642 decodes the
acquired DPCM data of the AC coefficients using the initial
settings and the like supplied from the initial setting unit
641, and determines AC coefficients. The AC coefficient
DPCM decoding unit 642 supplies the determined AC
coefficients (ScalingList[i]) to the inverse overlap
determination unit 553. The AC coefficient DPCM decoding
unit 642 executes the process described above on a plurality
of scaling lists.
[0356]
The DC coefficient DPCM decoding unit 643 acquires the
DPCM data of the DC coefficient
(scaling_list_delta_coef(dc)) supplied from the exp-G unit
551. The DC coefficient DPCM decoding unit 643 decodes the
127
acquired DPCM data of the DC coefficient using the initial
settings and the like supplied from the initial setting unit
641, and determines DC coefficients of the individual
scaling lists. The DC coefficient DPCM decoding unit 643
supplies the determined DC coefficients
(scaling_list_dc_coef) to the inverse overlap determination
unit 553.
[0357]
Accordingly, the inverse DPCM unit 552 can perform
correct DPCM decoding, and can obtain the DC coefficients
and the AC coefficients. That is, the image decoding device
300 can suppress an increase in the amount of coding of
scaling lists.
[0358]
<5-5. Flow of inverse DPCM process>
Also in the fourth method, the image decoding device
300 executes a quantization matrix decoding process in a
manner similar to that in the first method described above
with reference to the flowchart illustrated in Fig. 27.
Similarly, the image decoding device 300 executes a residual
signal decoding process in a manner similar to that in the
first method described above with reference to the flowchart
illustrated in Fig. 28.
[0359]
An example of the flow of an inverse DPCM process
128
executed by the inverse DPCM unit 552 will be described with
reference to a flowchart illustrated in Figs. 41 and 42.
[0360]
When the inverse DPCM process is started, the initial
setting unit 641 and the AC coefficient DPCM decoding unit
642 execute the processing of steps S511 to S517 in a manner
similar to that in the processing of steps S421 to S427 in
Fig. 34.
[0361]
If it is determined in step S514 that the variable i is
greater than or equal to coefNum, the AC coefficient DPCM
decoding unit 642 advances the process to step S518.
[0362]
In step S518, the AC coefficient DPCM decoding unit 642
determines whether or not all the scaling lists (difference
matrices) in which the DC coefficients are collectively
subjected to a DPCM process have been processed. If it is
determined that there is an unprocessed scaling list
(difference matrix), the AC coefficient DPCM decoding unit
642 returns the process to step S511, and repeatedly
performs the subsequent processing.
[0363]
Furthermore, if it is determined that there is no
unprocessed scaling list (difference matrix), the AC
coefficient DPCM decoding unit 642 advances the process to
129
Fig. 42.
[0364]
In step S521 in Fig. 42, the initial setting unit 641
sets sizeID and a variable nextcoef as follows.
sizeID = 2
nextcoef = 8
[0365]
Furthermore, in step S522, the initial setting unit 641
sets MatrixID as follows.
MatrixID = 0
[0366]
In step S523, the DC coefficient DPCM decoding unit 643
determines whether or not sizeID < 4. If it is determined
that sizeID is smaller than 4, the DC coefficient DPCM
decoding unit 643 advances the process to step S524.
[0367]
In step S524, the DC coefficient DPCM decoding unit 643
determines whether or not MatrixID < (sizeID == 3)?2:6 is
satisfied. If it is determined that MatrixID < (sizeID ==
3)?2:6 is satisfied, the DC coefficient DPCM decoding unit
643 advances the process to step S525.
[0368]
In step S525, the DC coefficient DPCM decoding unit 643
reads the DPCM data of the DC coefficient
(scaling_list_delta_coef).
130
[0369]
In step S526, the DC coefficient DPCM decoding unit 643
determines nextcoef as below using the read DPCM data, and
further determines scaling_dc_coef.
nextcoef = (nextcoef + scaling_list_delta_coef+256) %
256
scaling_dc_coef[sizeID - 2][MatrixID] = nextcoef
[0370]
In step S527, the DC coefficient DPCM decoding unit 643
increments MatrixID to change the processing target to the
subsequent DC coefficient (the subsequent scaling list or
residual matrix), and then returns the process to step S524.
[0371]
If it is determined in step S524 that MatrixID <
(sizeID == 3)?2:6 is not satisfied, the DC coefficient DPCM
decoding unit 643 advances the process to step S528.
[0372]
In step S528, the DC coefficient DPCM decoding unit 643
increments sizeID to change the processing target to the
subsequent DC coefficient (the subsequent scaling list or
residual matrix), and then returns the process to step S523.
[0373]
If it is determined in step S523 that sizeID is greater
than or equal to 4, the DC coefficient DPCM decoding unit
643 ends the inverse DPCM process, and returns the process
131
to Fig. 28.
[0374]
Accordingly, the differences between DC coefficients
can be correctly decoded. Therefore, the image decoding
device 300 can suppress an increase in the amount of coding
of scaling lists.
[0375]
<6. Sixth Embodiment
<6-1. Other syntax: First example>
Fig. 43 illustrates another example of the syntax for a
scaling list. This drawing corresponds to Fig. 12. In the
example illustrated in Fig. 12, the initial value of
nextcoef is set to a predetermined constant (for example, 8).
Alternatively, as illustrated in Fig. 43, the initial value
of nextcoef may be overwritten with the DPCM data of the DC
coefficient (scalinglistdccoefminus8).
[0376]
Accordingly, the amount of coding of the initial AC
coefficients (AC coefficients (0, 0)) in a 16x16 scaling
list and a 32x32 scaling list can be reduced.
[0377]
<6-2. Other syntax: Second example>
Fig. 44 illustrates another example of the syntax for a
scaling list. This drawing corresponds to Fig. 12.
[0378]
132
In the example illustrated in Fig. 12, when the value
of scaling_list_pred_matrix_id_delta, which is information
that specifies the reference destination in the copy mode,
is " 0 " , the scaling list that precedes the current scaling
list being processed by one scaling list is referred to, and
when the value of scaling_list_pred_matrix_id_delta is " 1 ",
the scaling list that precedes the current scaling list
being processed by two scaling lists is referred to.
[0379]
In contrast, in the example illustrated in Fig. 44, as
illustrated in part C of Fig. 44, when the value of
scaling_list_pred_matrix_id_delta, which is information that
specifies the reference destination in the copy mode, is " 0 ",
the default scaling list is referred to, and when the value
of scaling_list_pred_matrix_id_delta is " 1 " , the immediately
preceding scaling list is referred to.
[0380]
In this manner, modifying the semantics of
scaling_list_pred_matrix_id_delta can simplify the syntax in
a manner illustrated in part B of Fig. 44 and can reduce the
load of the DPCM process and the inverse DPCM process.
[0381]
<6-3. Other syntax: Third example>
Fig. 45 illustrates another example of the syntax for a
scaling list. This drawing corresponds to Fig. 12.
133
[0382]
In the example illustrated in Fig. 45, both of the
example illustrated in Fig. 43 and the example illustrated
in Fig. 44 described above are used.
[0383]
In the example illustrated in Fig. 45, accordingly, the
amount of coding of the initial AC coefficients (AC
coefficients (0, 0)) in a 16x16 scaling list and a 32x32
scaling list can be reduced. In addition, syntax can be
simplified and the load of the DPCM process and the inverse
DPCM process can be reduced.
[0384]
In the foregoing embodiments, the values of the
predetermined constants are arbitrary. In addition, the
sizes of the scaling lists are also arbitrary.
[0385]
Furthermore, while the foregoing description has been
given of a size transformation process for a scaling list, a
prediction matrix, or a difference matrix between them, the
size transformation process may be a process for actually
generating a matrix whose size has been transformed, or may
be a process for setting how to read each element in a
matrix from a memory (read control of matrix data) without
actually generating data of the matrix.
[0386]
134
In the size transformation process described above,
each element in a matrix whose size has been transformed is
constituted by any of the elements in the matrix whose size
has not yet been transformed. That is, a matrix whose size
has been transformed may be generated by reading elements in
a matrix whose size has not yet been transformed, which is
stored in a memory, using a certain method such as reading
some of the elements in the matrix or reading one element a
plurality of times.

We Claim:
1. An image processing device comprising:
a decoding unit configured to decode encoded data including an
initial difference coefficient that is a difference between a DC coefficient
and a predetermined initial value, a replacement difference coefficient that
is a difference between the DC coefficient and a (0, 0) coefficient located at
the beginning of an 8x8 quantization matrix, and further difference
coefficients each being a difference between two adjacent coefficients in a
sequence of coefficients of the 8x8 quantization matrix arranged in scan
order;
an inverse DPCM unit (552) configured to set the DC coefficient by
adding the initial value to the initial difference coefficient and to set the 8x8
quantization matrix by setting the (0, 0) coefficient located at the beginning
of the 8x8 quantization matrix by adding the DC coefficient to the
replacement difference coefficient and by setting the remaining coefficients
of the 8x8 quantization matrix by stepwise adding the further difference
coefficients to the (0, 0) coefficient, wherein a remaining coefficient at a
particular position in scan order is the sum of the (0, 0) coefficient and the
difference coefficient up to said particular position;
a matrix size transformation unit configured to upconvert the (8, 8)
quantization matrix into a 32x32 quantization matrix by performing a
nearest neighbor interpolation process on matrix elements of the 8x8
quantization matrix; and
a dequantization unit (440) configured to dequantize quantized data
obtained by decoding encoded data, using the DC coefficient and the 32x32
quantization matrix.
2. The image processing device as claimed in Claim 1, wherein
the decoding unit is configured to decode the encoded data including
a syntax in which the replacement difference coefficient and the difference
coefficients are collectively included as a difference coefficient group.
3. The image processing device as claimed in Claim 1, wherein
202
the replacement difference coefficient and the difference
coefficients are included as the syntax of the encoded data in the order of
the replacement difference coefficient and the difference coefficients, and
the decoding unit is configured to decode the replacement difference
coefficient and the difference coefficients in the order of the replacement
difference coefficient and the difference coefficients.
4. The image processing device as claimed in Claim 3, wherein
the initial difference value and the difference coefficient group are
included as the syntax of the encoded data in the order of the initial
difference value and the difference coefficient group, and the decoding unit
is configured to decode the initial difference value and the difference
coefficient group in the order of the initial difference value and the
difference coefficient group.
5. The image processing device as claimed in Claim 4, wherein
the initial difference value and the difference coefficient group are
included in the encoded data by performing an exponential golomb
encoding process, and
the decoding unit is configured to perform an exponential golomb
decoding process on the initial difference value and the difference
coefficient group obtained by performing the exponential golomb encoding
in the order of the initial difference value and the difference coefficient
group.
6. The image processing device as claimed in Claim 5, further comprising:
an inverse orthogonal transform unit (450) configured to inversely
transform the transform coefficient data generated by the dequantization
unit.
7. The image processing device as claimed in Claim 6, wherein
the inverse orthogonal transform unit (450) is configured to
inversely transform compared to the transforming performed by a 32x32
transform unit.
203
8. The image processing device as claimed in Claim 1, further comprising:
a replacement unit configured to replace a (0, 0) coefficient located
at the beginning of the 32x32 quantization matrix set by the matrix size
transformation unit with the DC coefficient set by the inverse DPCM unit;
9. The image processing device as claimed in Claim 8, wherein
the dequantization unit (440) is configured to dequantize quantized
data obtained by decoding encoded data, using the 32x32 quantization
matrix in which the (0, 0) coefficient located at the beginning has been
replaced with the DC coefficient by the replacement unit.
10. An image processing method comprising:
decoding encoded data including an initial difference coefficient that
is a difference between a DC coefficient and a predetermined initial value,
a replacement difference coefficient that is a difference between the DC
coefficient and a (0, 0) coefficient located at the beginning of an 8x8
quantization matrix, and further difference coefficients each being a
difference between two adjacent coefficients in a sequence of coefficients
of the 8x8 quantization matrix arranged in scan order;
setting the DC coefficient by adding the initial value to the initial
difference coefficient and setting the 8x8 quantization matrix by setting the
(0, 0) coefficient located at the beginning of the 8x8 quantization matrix by
adding the DC coefficient to the replacement difference coefficient and by
setting the remaining coefficients of the 8x8 quantization matrix by stepwise
adding the further difference coefficients to the (0, 0) coefficient, wherein a
remaining coefficient at a particular position in scan order is the sum of the
(0, 0) coefficient and the difference coefficient up to said particular position;
upconverting the (8, 8) quantization matrix into a 32x32
quantization matrix by performing a nearest neighbor interpolation process
on matrix elements of the 8x8 quantization matrix; and
dequantizing quantized data obtained by decoding encoded data,
using the DC coefficient and the 32x32 quantization matrix.
204
11. The image processing method as claimed in Claim 10, wherein
said decoding includes a syntax in which the replacement difference
coefficient and the difference coefficients are collectively included as a
difference coefficient group.
12. The image processing method as claimed in Claim 10, wherein
the replacement difference coefficient and the difference
coefficients are included as the syntax of the encoded data in the order of
the replacement difference coefficient and the difference coefficients, and
the replacement difference coefficient and the difference coefficients are
decoded in the order of the replacement difference coefficient and the
difference coefficients.
13. The image processing method as claimed in Claim 12, wherein
the initial difference value and the difference coefficient group are
included as the syntax of the encoded data in the order of the initial
difference value and the difference coefficient group, and the initial
difference value and the difference coefficient group are decoded in the
order of the initial difference value and the difference coefficient group.
14. The image processing method as claimed in Claim 13, wherein
the initial difference value and the difference coefficient group are
included in the encoded data by performing an exponential golomb
encoding process, and
the method comprises performing an exponential golomb decoding
process on the initial difference value and the difference coefficient group
obtained by performing the exponential golomb encoding in the order of the
initial difference value and the difference coefficient group.
15. The image processing method as claimed in Claim 14, further comprising:
inversely transforming the transform coefficient data generated by
said dequantization of the quantized data.
16. The image processing method as claimed in Claim 15, wherein
205
the transform coefficient data is inversely transformed compared to
the transforming performed by a 32x32 transform unit.
17. The image processing method as claimed in Claim 10, further comprising:
replacing a (0, 0) coefficient located at the beginning of the 32x32
quantization matrix set by said upconverting of the (8, 8) quantization
matrix with the DC coefficient set by adding the initial value to the initial
difference coefficient and to set the 8x8 quantization matrix by setting the
(0, 0) coefficient located at the beginning of the 8x8 quantization matrix by
adding the DC coefficient to the replacement difference coefficient and by
setting the remaining coefficients of the 8x8 quantization matrix by stepwise
adding the further difference coefficients to the (0, 0) coefficient.
18. The image processing method as claimed in Claim 17, further comprising:
dequantizing quantized data obtained by decoding encoded data,
using the 32x32 quantization matrix in which the (0, 0) coefficient located
at the beginning has been replaced with the DC coefficient.