1. A video decoding method, the method comprising: reconstructing a residual value by entropy-decoding received bitstream and dequantizing and inverse-transforming residual value information5 ; generating a final prediction unit by performing inter prediction on a prediction unit which is partitioned from a coding unit into at least two prediction units by asymmetric motion partitioning (AMP), the two partitioned prediction units comprising a first partitioned prediction unit and a second partitioned prediction unit; and 10 reconstructing a picture by adding the final prediction unit to the residual values, wherein the generating of the final prediction unit comprises performing interpolation using a filter tap with a variable length based on a horizontal length or a vertical length of the first partitioned prediction unit so that pixels in the second partitioned prediction unit are not involved in interpolation. 15

2. The method of claim 1, wherein the generating of the final prediction unit comprises performing interpolation for the first partitioned prediction unit and performing interpolation for the second partitioned prediction unit using filter taps with different lengths based on a horizontal length or a vertical length of the prediction units. 20

3. The method of claim 1, wherein the generating of the final prediction unit comprises performing horizontal interpolation for the first partitioned prediction unit using a horizontal filter tap shorter than a vertical filter tap when the first partitioned prediction unit is asymmetric and short in a horizontal direction. 25 31

4. The method of claim 1, wherein the generating of the final prediction unit comprises performing vertical interpolation for the first partitioned prediction unit using a vertical filter tap shorter than a horizontal filter tap when the first partitioned prediction unit is asymmetric and short in a vertical direction. 5

5. The method of claim 1, wherein the generating of the final prediction unit comprises performing horizontal interpolation for the first partitioned prediction unit using a horizontally short filter tap shorter than a filter tap for the second partitioned prediction unit which is horizontally long when the first partitioned prediction unit is asymmetric and 10 short in a horizontal direction.

6. The method of claim 1, wherein the generating of the final prediction unit comprises performing vertical interpolation for the first partitioned prediction unit using a vertically short filter tap shorter than a filter tap for the second partitioned prediction unit 15 which vertically long when the first partitioned prediction unit is asymmetric and short in a vertical direction.

7. The method of claim 1, wherein a 4-tap filter is used for the first partitioned prediction unit in vertical interpolation and a 6-tap filter is used for the second 20 partitioned prediction unit in vertical interpolation when a 64 x 64 unit to be predicted is asymmetrically partitioned in a vertical direction into 2N x nU or 2N x nD prediction units, N being a natural number, a 2N x nU prediction unit being a partitioned form having an upper block with a smaller area, and a 2N x nD prediction unit being a partitioned form having a lower block with a smaller area, and a 4-tap filter is used for the first partitioned 25 prediction unit in horizontal interpolation and a 6-tap filter is used for the second 32 partitioned prediction unit in horizontal interpolation when the 64 x 64 unit to be predicted is asymmetrically partitioned in a horizontal direction into nL x 2N or nR x 2N prediction units, N being a natural number, an nL x 2N prediction unit being a partitioned form having a left block with a smaller area, and an nR x 2N prediction unit being a partitioned form having a right block with a smaller area5 .

8. The method of claim 1, wherein a total length of filter taps in an asymmetric direction of the first and the second partitioned prediction units is larger than a length of a filter tap in a direction other than the asymmetric direction. 10

9. The method of claim 1, wherein the received bitstream comprises information on a prediction mode and a form of a prediction unit corresponding to a decoding target block. 15 10. The method of claim 9, wherein the received bitstream further comprises information on a length of an interpolation filter tap of the prediction unit corresponding to the decoding target block.

11. The method of claim 1, wherein the generating of the final prediction unit 20 comprises obtaining, from the bitstream, partition information on which direction the partitioned prediction units are asymmetrical; determining, based on the partition information, which asymmetric direction the partitioned prediction units have a longer length; determining a length of a filter tap to be used for interpolation based on a determination result; and performing interpolation using the determined filter tap. 25 33

12. A video decoding apparatus, the apparatus comprising: a residual value reconstructing module to reconstruct a residual value by entropydecoding a received bitstream and dequantizing and inverse-transforming residual value information; a final prediction unit generating module to generate a final prediction unit b5 y performing inter prediction on a prediction unit which is a block partitioned from a coding unit into at least two prediction units by asymmetric motion partitioning (AMP), the two partitioned prediction units comprising a first partitioned prediction unit and a second partitioned prediction unit; and 10 a picture reconstructing module to reconstruct a picture by adding the final prediction unit and the residual value, wherein the final prediction unit generating module performs interpolation using a filter tap with a variable length based on a horizontal length or a vertical length of the first partitioned prediction unit so that pixels in the second partitioned prediction unit are not 15 involved in interpolation.

13. A video encoding method, the method comprising: performing inter prediction on a prediction unit obtained by partitioning an input picture using asymmetric motion partitioning (AMP) to predict and encode the picture, the 20 partitioned prediction unit comprising a first partitioned prediction unit and a second partitioned prediction unit; and transforming and quantizing a residual value that is a difference between a prediction unit generated by the inter prediction and a current prediction unit, and entropyencoding thereon, 25 wherein the performing of the inter prediction comprises performing interpolation 34 using a filter tap with a variable length based on a horizontal length or a vertical length of the first partitioned prediction unit so that pixels in the second partitioned prediction unit are not involved in interpolation.

14. The method of claim 13, wherein the performing of the inter predictio5 n comprises performing interpolation for the first partitioned prediction unit and performing interpolation for the second partitioned prediction unit using filter taps with different lengths based on a horizontal length or a vertical length of the prediction units. 10 15. The method of claim 13, wherein the performing of the inter prediction comprises performing horizontal interpolation for the first partitioned prediction unit using a horizontal filter tap shorter than a vertical filter tap when the first partitioned prediction unit is asymmetric and short in a horizontal direction. 15 16. The method of claim 13, wherein the performing of the inter prediction comprises performing horizontal interpolation using a horizontally short filter tap shorter than a filter tap for the second partitioned prediction unit which is horizontally long when the first partitioned prediction unit is asymmetric and short in a horizontal direction. 20 17. The method of claim 13, wherein a total length of filter taps in an asymmetric direction of the first and the second partitioned prediction units is larger than a length of a filter tap in a direction other than the asymmetric direction.

18. The method of claim 13, wherein the performing of the inter prediction 25 comprises acquiring information on which direction the partitioned prediction units are 35 asymmetrical; determining, based on the obtained information, which asymmetric direction the partitioned prediction units have a longer length; determining a length of a filter tap to be used for interpolation based on a determination result; and performing interpolation using the determined filter tap. 5

19. The method of claim 13, wherein transforming and quantizing a residual value, and entropy-encoding thereon comprises generating a bitstream, and wherein the bitstream comprises information on a length of an interpolation filter tap of the prediction unit corresponding to an encoding target block. 10

20. A video encoding apparatus, the appratus comprising: an inter prediction module to perform inter prediction on a prediction unit obtained by partitioning an input picture using asymmetric motion partitioning (AMP) to predict and encode the picture, the partitioned prediction unit comprising a first partitioned 15 prediction unit and a second partitioned prediction unit; and an entropy encoding module to entropy-encode a residual value which is transformed and/or quantized, wherein the residual value is a difference between a prediction unit generated by the inter prediction and a current prediction unit, wherein the inter prediction module performs interpolation using a filter tap with a 20 variable length based on a horizontal length or a vertical length of the first partitioned 36 prediction unit so that pixels in the second partitioned prediction unit are not involved in interpolation.

【DESCRIPTION】
【Technical Field】
[1] The present invention relates to image processing, and more particularly, to an
inter prediction method and an inter prediction apparatus5 .
【Background Art】
[2] Recently, demands for high-resolution and high-quality videos, such as highdefinition
(HD) and ultrahigh-definition (UHD) videos, are increasing.
[3] To provide videos with higher resolution and higher quality, the amount of
10 video data increases. Accordingly, costs of transferring and storing video data rise so as
to provide high-quality videos as compared with conventional video data processing
methods. In order to solve these problems occurring with an increase in resolution and
quality of video data, high-efficiency video compression techniques may be utilized.
[4] As video data compression technology, various schemes are used such as inter
15 prediction of predicting pixel values included in a current picture from other pictures, intra
prediction of predicting pixel values included in a current picture using information on
other pixels than the current picture, and entropy encoding/decoding of allocating shorter
codes to frequently occurring or appearing signals.
【Disclosure】
20 【Technical Problem】
[5] An aspect of the present invention is to provide a video encoding method and
a video encoding apparatus which are capable of increasing video encoding performance.
[6] Another aspect of the present invention is to provide a video decoding method
and a video decoding apparatus which are capable of increasing video decoding
25 performance.
3
[7] Still another aspect of the present invention is to provide an inter encoding
method and an inter encoding apparatus which are capable of increasing video encoding
performance.
【Technical Solution】
[8] An embodiment of the present invention provides a video decoding metho5 d
including reconstructing a residual value by entropy-decoding a received bitstream and
dequantizing and inverse-transforming residual value information, generating a final
prediction unit by performing inter prediction on a prediction unit which is partitioned
form a coding unit into at least two prediction units by asymmetric motion partitioning
10 (AMP), the two partitioned prediction units including a first partitioned prediction unit and
a second partitioned prediction unit, and reconstructing a picture by adding the final
prediction unit to the residual value, wherein the generating of the final prediction unit
includes performing interpolation using a filter tap with a variable length based on a
horizontal length or a vertical length of the first partitioned prediction unit so that pixels in
15 the second partitioned prediction unit are not involved in interpolation.
[9] The generating of the final prediction unit may include performing
interpolation for the first partitioned prediction unit and performing interpolation for the
second partitioned prediction unit using filter taps with different lengths based on a
horizontal length or a vertical length of the prediction unit.
20 [10] The generating of the final prediction unit may include performing
horizontal interpolation for the first partitioned prediction unit using a horizontal filter tap
shorter than a vertical filter tap when the first partitioned prediction unit is asymmetric and
short in a horizontal direction.
[11] The generating of the final prediction unit may include performing vertical
25 interpolation for the first partitioned prediction unit using a vertical filter tap shorter than a
4
horizontal filter tap when the first partitioned prediction unit is asymmetric and short in a
vertical direction.
[12] The generating of the final prediction unit may include performing
horizontal interpolation for the first partitioned prediction unit using a horizontally short
filter tap shorter than a filter tap for the second partitioned prediction unit which i5 s
horizontally long when the first partitioned prediction unit is asymmetric and short in a
horizontal direction.
[13] The generating of the final prediction unit may include performing vertical
interpolation for the first partitioned prediction unit using a vertically short filter tap
10 shorter than a filter tap for the second partitioned prediction unit which is vertically long
when the first partitioned prediction unit is asymmetric and short in a vertical direction.
[14] A 4-tap filter may be used for the first partitioned prediction unit in vertical
interpolation and a 6-tap filter is used for the second partitioned prediction unit in vertical
interpolation when a 64 x 64 unit to be predicted is asymmetrically partitioned in a vertical
15 direction into 2N x nU or 2N x nD prediction units, N being a natural number, a 2N x nU
prediction unit being a partitioned form having an upper block with a smaller area, and a
2N x nD prediction unit being a partitioned form having a lower block with a smaller area,
and a 4-tap filter may be used for the first partitioned prediction unit in horizontal
interpolation and a 6-tap filter is used for the second partitioned prediction unit in
20 horizontal interpolation when the 64 x 64 unit to be predicted is asymmetrically
partitioned in a horizontal direction into nL x 2N or nR x 2N prediction units, N being a
natural number, an nL x 2N prediction unit being a partitioned form having a left block
with a smaller area, and an nR x 2N prediction unit being a partitioned form having a right
block with a smaller area.
25 [15] A total length of filter taps in an asymmetric direction of the first and the
5
second partitioned prediction units may be larger than a length of a filter tap in a direction
other than the asymmetric direction.
[16] The received bitstream may include information on a prediction mode and a
form of a prediction unit corresponding to the decoding target block.
[17] The received bitstream may further include information on a length of a5 n
interpolation filter tap of the prediction unit corresponding to the decoding target block.
[18] The generating of the final prediction unit may include acquiring, from the
bitstream, partition information on which direction the partitioned prediction units are
asymmetrical; determining, based on the partition information, which asymmetric
10 direction the partitioned prediction units have a longer length; determining a length of a
filter tap to be used for interpolation based on a determination result; and performing
interpolation using the determined filter tap.
[19] Another embodiment of the present invention provides a video decoding
apparatus including a residual value reconstructing module to reconstruct a residual value
15 by entropy-decoding a received bitstream and dequantizing and inverse-transforming
residual value information, a final prediction unit generating module to generate a final
prediction unit by performing inter prediction on a prediction unit which is partitioned
from a coding unit into at least two prediction units by AMP, the two partitioned
prediction units including a first partitioned prediction unit and a second partitioned
20 prediction unit; and a picture reconstructing module to reconstruct a picture by adding the
final prediction unit to the residual value, wherein the final prediction unit generating
module performs interpolation using a filter tap with a variable length based on a
horizontal length or a vertical length of the first partitioned prediction unit so that pixels in
the second partitioned prediction unit are not involved in interpolation.
25 [20] Still another embodiment of the present invention provides a video encoding
6
method including performing inter prediction on a prediction unit obtained by partitioning
an input picture using AMP to predict and encode the picture, the partitioned prediction
unit including a first partitioned prediction unit and a second partitioned prediction unit,
and transforming and quantizing a residual value that is a difference between a prediction
unit generated by the inter prediction and a current prediction unit, and entropy-encodin5 g
thereon, wherein the performing of the inter prediction includes performing interpolation
using a filter tap with a variable length based on a horizontal length or a vertical length of
the first partitioned prediction unit so that pixels in the second partitioned prediction unit
are not involved in interpolation.
10 [21] The performing of the inter prediction may include performing interpolation
for the first partitioned prediction unit and performing interpolation for the second
partitioned prediction unit using filter taps with different lengths based on a horizontal
length or a vertical length of the prediction units.
[22] The performing of the inter prediction may include performing horizontal
15 interpolation for the first partitioned prediction unit using a horizontal filter tap shorter
than a vertical filter tap when the first partitioned prediction unit is asymmetric and short
in a horizontal direction.
[23] The performing of the inter prediction may include performing horizontal
interpolation using a horizontally short filter tap shorter than a filter tap for the second
20 partitioned prediction unit which is horizontally long when the first partitioned prediction
unit is asymmetric and short in a horizontal direction.
[24] A total length of filter taps in an asymmetric direction of the first and the
second partitioned prediction units may be larger than a length of a filter tap in a direction
other than the asymmetric direction.
25 [25] The performing of the inter prediction may include acquiring information on
7
which direction the partitioned prediction units are asymmetrical; determining, based on
the obtained information, which asymmetric direction the partitioned prediction units have
a longer length; determining a length of a filter tap to be used for interpolation based on a
determination result; and performing interpolation using the determined filter tap.
[26] The transforming and quantizing a residual value, and entropy-encodin5 g
thereon comprises generating a bitstream, and wherein the bitstream comprises
information on a length of an interpolation filter tap of the prediction unit corresponding
to an encoding target block
[27] Yet another embodiment of the present invention provides a video encoding
10 apparatus including an inter prediction module to perform inter prediction on a prediction
unit obtained by partitioning an input picture using AMP to predict and encode the picture,
the partitioned prediction unit including a first partitioned prediction unit and a second
partitioned prediction unit, and an entropy encoding module to entropy-encode a residual
value which is transformed and/or quantized, wherein the residual value is a difference
15 between a prediction unit generated by the inter prediction and a current prediction unit,
and wherein the inter prediction module performs interpolation using a filter tap with a
variable length based on a horizontal length or a vertical length of the first partitioned
prediction unit so that pixels in the second partitioned prediction unit are not involved in
interpolation.
20 【Advantageous Effects】
[28] According to a video encoding method and a video encoding apparatus of the
present invention, video encoding performance may be enhanced.
[29] According to a video decoding method and a video decoding apparatus of the
present invention, video decoding performance may be enhanced.
25 [30] According to an inter prediction encoding method and an inter prediction
8
encoding apparatus of the present invention, video encoding/decoding performance may
be enhanced.
【Description of Drawings】
[31] FIG. 1 is a block diagram illustrating a configuration of a video encoding
apparatus according to an exemplary embodiment of the present invention5 .
[32] FIG. 2 is a block diagram illustrating a configuration of a video decoding
apparatus according to an exemplary embodiment of the present invention.
[33] FIG. 3 schematically illustrates interpolation in inter prediction according to
an exemplary embodiment of the present invention.
10 [34] FIG. 4 schematically illustrates that an interpolation filter tap is used when
asymmetric motion partitioning (AMP) is used in a vertical direction in the video
encoding apparatus according to an exemplary embodiment of the present invention.
[35] FIG. 5 schematically illustrates that an interpolation filter tap is used when
AMP is used in a horizontal direction in the video encoding apparatus according to an
15 exemplary embodiment of the present invention.
[36] FIG. 6 is a flowchart schematically illustrating a process of performing inter
prediction for an asymmetrically partitioned PU in the video encoding apparatus according
to an exemplary embodiment of the present invention.
[37] FIG. 7 schematically illustrates that an interpolation filter tap adequate for a
20 vertical or horizontal length of a partitioned PU is used when AMP is used in the vertical
direction in the video encoding apparatus according to an exemplary embodiment of the
present invention.
[38] FIG. 8 schematically illustrates that an interpolation filter tap adequate for a
vertical or horizontal length of a partitioned PU is used when AMP is used in the
25 horizontal direction in the video encoding apparatus according to an exemplary
9
embodiment of the present invention.
[39] FIG. 9 is a flowchart schematically illustrating a video encoding method
according to an exemplary embodiment of the present invention.
[40] FIG. 10 is a flowchart schematically illustrating a video decoding method
according to an exemplary embodiment of the present invention5 .
【Mode for Invention】
[41] The present invention may be changed and modified variously and be
illustrated with reference to different exemplary embodiments, some of which will be
described and shown in the drawings.
10 [42] However, these embodiments are not intended for limiting the invention but
are construed as including includes all modifications, equivalents and replacements which
belong to the spirit and technical scope of the invention.
[43] Although the terms first, second, etc. may be used to describe various
elements, these elements should not be limited by these terms. These terms are used only
15 to distinguish one element from another element. For example, a first element could be
termed a second element and a second element could be termed a first element likewise
without departing from the teachings of the present invention. The term "and/or"
includes any and all combinations of a plurality of associated listed items.
[44] It will be understood that when an element is referred to as being "connected
20 to" or “coupled to" another element, the element can be directly connected or coupled to
another element or intervening elements. On the contrary, when an element is referred to
as being "directly connected to" or "directly coupled to" another element, there are no
intervening elements present.
[45] The terminology used herein is for the purpose of describing particular
25 embodiments only and is not intended to be limiting of the invention. As used herein,
10
the singular forms "a," "an" and "the" are intended to include the plural forms as well,
unless the context clearly indicates otherwise. It will be further understood that the terms
"include" and/or "have," when used in this specification, specify the presence of stated
features, integers, steps, operations, elements, and/or components, but do not preclude the
presence or addition of one or more other features, integers, steps, operations, elements5 ,
components, and/or groups thereof.
[46] Unless otherwise defined, all terms (including technical and scientific terms)
used herein have the same meaning as commonly understood by one of ordinary skill in
the art to which this invention belongs. It will be further understood that terms, such as
10 those defined in commonly used dictionaries, should be interpreted as having a meaning
that is consistent with their meaning in the context of the relevant art and will not be
interpreted in an idealized or overly formal sense unless expressly so defined herein.
[47] Hereinafter, exemplary embodiments of the invention will be described in
detail with reference to the accompanying drawings. For ease of understanding, like
15 reference numerals in the drawings refer to like elements throughout, and redundant
descriptions of like elements will be omitted herein.
[48]
[49] FIG. 1 is a block diagram illustrating a configuration of a video encoding
apparatus according to an exemplary embodiment of the present invention. Referring to
20 FIG. 1, the video encoding apparatus may include a picture partition module 110, an inter
prediction module 120, an intra prediction module 125, a transform module 130, a
quantization module 135, a dequantization module 140, an inverse transform module 145,
a module filter 150, a memory 155, a rearrangement module 160 and an entropy encoding
module 165.
25 [50] The picture partition module 110 may divide an input picture into one or
11
more coding units. A coding unit (CU) is a unit of encoding performed by the video
encoding apparatus and may be recursively split with depth information based on a
quadtree structure. A CU may have different sizes of 8 x 8, 16 x 16, 32 x 32 and 64 x 64.
A CU with a maximum size is referred to as a largest coding unit (LCU), and a CU with a
minimum size as a smallest coding unit (SCU)5 .
[51] The picture partition module 110 may divide a CU to generate a prediction
unit (PU) and the picture partition module 110 may divide a CU to generate a transform
unit (TU). A PU may be smaller than or the same as a CU, and may not necessarily be a
square block but be a rectangular block.
10 [52] Generally, intra prediction may be performed by 2N*2N or N*N blocks.
Here, N is a natural number, representing a number of pixels, and 2N*2N or N*N may
represent a PU size (and/or partition mode). However, in short distance intra prediction
(SDIP), not only a 2N*2N PU but a further subdivided PU with a size of hN*2N/2N*hN
(here, h=1/2) may be also used to increase efficiency in intra prediction. When an
15 hN*2N PU or 2N*hN PU is used, directionality of a boundary in a block may be reflected
well, and accordingly energy of a prediction error signal may be decreased to reduce an
amount of bits needed for encoding, thereby increasing encoding efficiency.
[53] Inter prediction may be performed by 2N*2N, 2N*N, N*2N or N*N blocks.
Here, N is a natural number, representing a number of pixels, and 2N*2N, 2N*N, N*2N
20 or N*N may represent a PU size (and/or partition mode). Further, inter prediction may
be performed in the unit of 2NxnU PU, 2NxnD PU, nLx2N PU or nRx2N PU, in addition
to the 2N*2N PU, 2N*N PU, N*2N PU or N*N PU, in order to enhance efficiency in inter
prediction. Here, 2NxnU, 2NxnD, nLx2N or nRx2N may represent a PU size (and/or
partition mode). In 2NxnU and 2NxnD partition modes, a PU may have a size of
25 2Nx(1/2)N or 2Nx(3/2)N, while in nLx2N and nRx2N partition modes, a PU may have a
12
size of (1/2)Nx2N or (3/2)Nx2N.
[54] In an inter prediction mode, the inter prediction module 120 may perform
motion estimation (ME) and motion compensation (MC). The inter prediction module
120 may generate a prediction block based on information on at least one of previous and
subsequent pictures of the current picture5 .
[55] The inter prediction module 120 may perform motion estimation based on a
split target prediction block and at least one reference block stored in the memory 155.
The inter prediction module 120 may generate motion information including a motion
vector (MV), a reference block index and a prediction mode, etc. as a result of motion
10 estimation.
[56] Further, the inter prediction module 120 may perform motion compensation
using the motion information and the reference block. Here, the inter prediction module
120 may generate a prediction block from the reference block corresponding to an input
block and output the predicted block.
15 [57] In an intra prediction mode, the intra prediction module 125 may generate a
prediction block based on information on a pixel in the current picture. In the intra
prediction mode, the intra prediction module 125 may perform prediction for a current
block based on a target prediction block and a reconstructed block previously
reconstructed after transform and quantization. Here, the reconstructed block may be a
20 reconstructed picture that has not been subjected to the filter module 150.
[58] In the inter prediction mode or intra prediction mode described above,
prediction may be performed on a prediction target block to generate a prediction block.
Here, a residual block may be generated based on a differential value between prediction
target block (original block) and the generated prediction block.
25 [59] The transform module 130 may transform a residual block by a TU to
13
generate a transform coefficient. A TU may have a tree structure within maximum and
minimum sizes. It may be indicated through a flag whether a current block is split into
sub-blocks by each TU. The transform module 130 may perform transform based on a
discrete cosine transform (DCT) and/or discrete sine transform (DST).
[60] The quantization module 135 may quantize values transformed by th5 e
transform module 130. A quantization coefficient may change based on a block or
priority of a picture. The quantized transform coefficient may be provided to the
rearrangement module 160 and the dequantization module 140.
[61] The rearrangement module 160 may arrange a two-dimensional block of the
10 quantized transform coefficients into a one-dimensional vector of transform coefficients
by scanning so as to enhance efficiency in entropy encoding. The rearrangement module
160 may change a scanning order based on stochastic statistics to enhance entropy
encoding efficiency.
[62] The entropy encoding module 165 may entropy-encode the values obtained
15 by the rearrangement module 160. In entropy encoding, a codeword with smaller
number of bits may be allocated to a more frequently occurring syntax element value,
while a codeword with more number of bits may be allocated to a less frequently
occurring syntax element value. Thus, a size of a bit string for symbols to be encoded
may be reduced to enhance video encoding compression performance. Various encoding
20 methods, such as exponential Golomb coding, context-adaptive variable length coding
(CAVLC) and/or context-adaptive binary arithmetic coding (CABAC), may be used for
entropy encoding. The encoded information may be formed into a compressed bitstream
and be transferred or stored through a network abstraction layer (NAL).
[63] The dequantization module 140 may dequantize the transform coefficients
25 quantized by the quantization module 135, and the inverse transform module 145 may
14
inverse-transform the dequantized transform coefficients to generate a reconstructed
residual block. The reconstructed residual block may be added to the prediction block
generated by the inter prediction module 120 or the intra prediction module 125 to
generate a reconstructed block. The reconstructed block may be provided to the intra
prediction module 125 and the filter module 1505 .
[64] The filter module 150 may perform a deblocking filter, a sample adaptive
offset (SAO) and/or an adaptive loop filter (ALF) on the reconstructed residual block.
Further, the deblocking filter may perform filtering on the reconstructed block so as to
remove a distortion on boundaries between blocks occurring in encoding and decoding.
10 The SAO is a loop filtering process to be performed on the block, for which the
deblocking filtering process is completed, to reduce the difference from an original picture
by a pixel. A band offset and an edge offset may be used as the SAO. The band offset
may divide a pixel into 32 bands according to intensity and apply offsets to two divided
groups of 16 bands on an edge area and 16 bands in a central area. The ALF may
15 perform filtering so as to minimize an error between the target prediction block and the
finally reconstructed block. The ALF may perform filtering based on a value obtained
by comparing the reconstructed block filtered by the deblocking filter with the current
target prediction block, and filter coefficient information on the ALF may be signaled in a
slice header from the encoding apparatus to the decoding apparatus.
20 [65] The memory 155 may store the finally reconstructed block via the filter
module 150, and the finally reconstructed block may be provided to the inter prediction
module 120 performing inter prediction.
[66]
[67] FIG. 2 is a block diagram illustrating a configuration of a video decoding
25 apparatus according to an exemplary embodiment of the present invention. Referring to
15
FIG. 2, the video decoding apparatus may include an entropy decoding module 210, a
rearrangement module 215, a dequantization module 220, an inverse transform module
225, an inter prediction module 230, an intra prediction module 235, a filter module 240
and a memory 245.
[68] The entropy decoding module 210 may obtain a compressed bitstream fro5 m
an NAL. The entropy decoding module 210 may entropy-decode the obtained bitstream,
and also entropy-decode a prediction mode information and motion vector information if
the bitstream includes the prediction mode information and the motion vector information.
When entropy decoding is used, a codeword with smaller number of bits may be allocated
10 to a more frequently occurring syntax element value, while a codeword with more number
of bits may be allocated to a less frequently occurring syntax element value. Thus, a size
of a bit string for symbols to be encoded may be reduced to enhance video decoding
performance.
[69] An entropy-decoded transform coefficient or residual signal may be provided
15 to the rearrangement module 215. The rearrangement module 215 may inverse-scan the
decoded transform coefficient or residual signal to generate a 2D block of transform
coefficients.
[70] The dequantization module 220 may dequantize the rearranged transform
coefficients. The inverse transform module 225 may inverse-transform the dequantized
20 transform coefficients to generate a residual block.
[71] The residual block may be added to a prediction block generated by the inter
prediction module 230 or intra prediction module 235 to generate a reconstructed block.
The reconstructed block may be provided to the intra prediction module 235 and the filter
module 240. The inter prediction module 230 and the intra prediction module 235
25 performs operations the same as or the equivalent to those of the inter prediction module
16
120 and the intra prediction module 125 of the video encoding apparatus, and thus
descriptions thereof will be omitted herein.
[72] The filter module 240 may perform filtering on the reconstructed block using
a deblocking filter, an SAO and/or an ALF. The deblocking filter may perform filtering
on the reconstructed blocks to remove a distortion on a boundary between blocks tha5 t
occurs in encoding and decoding. The SAO may be applied to the reconstructed block,
for which the deblocking filtering process is completed, to reduce a difference from an
original picture by a pixel. The ALF may perform filtering on the reconstructed block
for which the SAO is completed, so as to minimize an error between the target prediction
10 block and the finally reconstructed block.
[73] The memory 245 may store the finally reconstructed block obtained through
the filter module 240, and the stored reconstructed block may be provided to the inter
prediction module 230 performing inter prediction.
[74] Hereinafter, a block may refer to a video encoding and decoding unit. Thus,
15 in this specification, a block may mean a CU, PU or TU. Also, a encoding/decoding
target block may collectively include a transform/inverse transform target block, if
transform/inverse transform is performed, and a target prediction block, if prediction is
performed.
[75]
20 [76] FIG. 3 schematically illustrates interpolation in inter prediction according to
an exemplary embodiment of the present invention. As shown in FIG. 3, when the
encoding apparatus (and/or decoding apparatus) generates a signal of a PU using motion
information on inter prediction, an 8-tap interpolation filter may be used.
[77] Referring to FIG. 3, interpolation is performed for each location in a
25 horizontal direction or vertical direction to predict a pixel value (including luma and
17
chroma values). As described above, using the 8-tap interpolation filter means that if a
PU is a predetermined 4 x 4 block (e.g. indicating a current block 310), eight pixel values
in the right direction and the left direction (i.e. in the horizontal direction) with respect to
the 4 x 4 block or in the upward direction and downward direction (i.e. in the vertical
direction) with respect to the 4 x 4 block are properly used in interpolation to predict pixe5 l
values of the current block 310. Although FIG. 3 illustrates usage of the 8-tap filter only,
the present invention is not limited thereto.
[78] In the present embodiment, 8-tap interpolation may be performed in the
horizontal direction and then 8-tap interpolation may be performed in the vertical direction.
10 First, assuming that a pixel value of a top left pixel of each 4 x 4 block is known, a pixel
value of a pixel (a0,0) just to the right of the top left pixel may be predicted by
interpolation using pixel values of top left pixels of three 4 x 4 blocks on the left of the
current block and pixel values of top left pixels of four 4 x 4 blocks on the right of the
current block, which is expressed by the following equation.
15 [79] [Equation 1]
[80] Here, shift1=BitDepthY (Bit depth of Y component) - 8. In this way, pixel
values of other pixels in the current block 310 may be predicted by interpolation, which is
expressed by the following equation.
20 [81] [Equation 2]
18
[82] Here, shift2=8. As shown in Equation 2, pixel values of three upper pixels
and three left pixels other than the top left pixel of the current block 310 may be predicted
by horizontal or vertical interpolation using pixel values of top left pixels of vertically or
horizontally neighboring 4 x 4 blocks, and pixel values of remaining pixels may b5 e
predicted by vertical or horizontal interpolation using pixels values of upper pixels of
seven 4 x 4 blocks vertically or horizontally neighboring. Using Equation 1 or 2, pixel
values of a PU to be currently predicted may be derived and a prediction signal related to
the PU may be generated.
10 [83] FIG. 4 schematically illustrates that an interpolation filter tap is used when
asymmetric motion partitioning (AMP) is used as asymmetric in the vertical direction in
the video encoding apparatus according to an exemplary embodiment of the present
invention.
[84] Referring to FIG. 4, when a PU is partitioned as AMP and a long filter tap,
19
such as an 8-tap, is used for a shorter direction of an asymmetrical partition, pixels of a
different partition are also involved in the interpolation. In this case, pixels have a weak
correlation when belonging to different partitions, and thus interpolation efficiency is
likely to decrease. That is, when the block 412 and the block 414 are interpolated
together, interpolation efficiency decreases due to a weak correlation between the bloc5 k
412 and the block 414. The same result is brought to the block 422 and the block 424.
[85] According to the present embodiment, when a PU is partitioned as AMP, a
smaller filter tap than a conventional filter tap may be used for interpolation in an
asymmetric direction of an asymmetrically partitioned PU with a shorter length. For
10 example, a smaller filter tap than an 8-tap may be used to perform interpolation of the
asymmetrically partitioned PU with the shorter length. In an inter mode, a PU may have
a 2N*2N, 2N*N, N*2N, N*N, 2NxnU, 2NxnD, nLx2N or nRx2N form. An 8-tap filter
may be used for interpolation of a symmetrically partitioned PU, such as 2N*2N, 2N*N,
N*2N and N*N PUs.
15 [86] Referring to left illustration of FIG. 4, when the PU is partitioned as a shape
of 2N x nU block 410, which is asymmetrically partitioned in the vertical direction, an
upper block 412 is a partitioned block with a shorter length. The block 410 may include
the upper block 412 and a lower block 414, in which a ratio between lengths in the vertical
direction of the upper block 412 and the lower block 414 may be 16: 48. Referring to a
20 right illustration of FIG. 4, when the PU is partitioned as a shape of a 2N x nD block 420,
which is asymmetrically partitioned in the vertical direction, a lower block 424 is a
partitioned block with a shorter length. The block 420 may include an upper block 422
and the lower block 424, in which a ratio between lengths in the vertical direction of the
upper block 422 and the lower block 424 may be 48: 16. When asymmetrically
25 partitioned in the vertical direction, the upper block 412 of the 2N x nU block 410 and the
20
lower block 424 of the 2N x nD block 420 may be interpolated using a smaller tap in the
vertical direction than in the horizontal direction. For example, when an 8-tap filter is
used in the horizontal direction, a filter with a smaller tap than an 8-tap may be used in the
vertical direction.
[87] FIG. 5 schematically illustrates that an interpolation filter tap is used whe5 n
AMP is used as asymmetric in the horizontal direction in the video encoding apparatus
according to an exemplary embodiment of the present invention.
[88] Referring to a left illustration of FIG. 5, when the PU is partitioned as a
shape of an nL x 2N block 510, which is asymmetrically partitioned in the horizontal
10 direction, a left block 512 is a partitioned block with a shorter length. The block 510
may include the left block 512 and a right block 514, in which lengths in the vertical
direction of the left block 512 and the right block 514 may be the same that is 64 while a
ratio between lengths thereof in the horizontal direction may be 16: 48. Referring to a
right illustration of FIG. 5, when the PU is partitioned as a shape of a nR x 2N block 520,
15 which is asymmetrically partitioned in the horizontally direction, a right block 524 is a
partitioned block with a shorter length. The block 520 may include a left block 522 and
the right block 524, in which a ratio between lengths in the horizontal direction of the left
block 522 and the right block 524 may be 48: 16. When asymmetrically partitioned in
the horizontal direction, the left block 512 of the nL x 2N block 510 and the right block
20 524 of the nR x 2N block 520 may be interpolated using a smaller tap in the horizontal
direction than in the vertical direction. For example, a filter with a smaller tap than an 8-
tap may be used in the horizontal direction.
[89] Although exemplary embodiments of FIGS. 4 and 5 have been described
with reference to a 64 x 64 block, the exemplary embodiments also can be applied for
25 blocks with various sizes or shapes other than a 64 x 64 block.
21
[90] FIG. 6 is a flowchart schematically illustrating a process of performing inter
prediction for an asymmetrically partitioned PU in the video encoding apparatus according
to an exemplary embodiment of the present invention. As shown in FIG. 6, the process
of performing inter prediction may include obtaining partition information (S610),
determining a length in an asymmetric direction (S620), determining a length of a filte5 r
tap (S630) and performing interpolation (S640).
[91] Referring to FIG. 6, in obtaining the partition information (S610), partition
information on an asymmetrically partitioned block is obtained. In the encoding process,
the partition information may be included in motion information on a current PU through
10 motion estimation. The motion information may include information on a motion vector
of the PU, a reference picture index, a prediction direction index, a prediction mode and a
information on the shape of the PU.
[92] According to the present embodiment, since a bitstream may be generated
including information on a length of an interpolation filter tap of the PU corresponding to
15 an encoding target block in the encoding process, the decoding apparatus may obtain
information on the length of the interpolation filter tap of the PU corresponding to a
decoding target block from the received bitstream. In this case, determining the length
(S620) and determining the length of the filter tap (S630) may be omitted. When the
bitstream does not include the information on the length of the filter tap, the information
20 on the shape of the PU may be obtained, followed by determining the length (S620) and
determining the length of the filter tap (S630), thereby determining the length of the filter
tap.
[93] In determining the length in the asymmetric direction (S620), the encoding
apparatus (and/or decoding apparatus) determines a length in the asymmetric direction
25 (either in the vertical direction or in the horizontal direction) of the PU corresponding to
22
the encoding (and/or decoding) target block based on the obtained partition information.
That is, the encoding apparatus determines whether an asymmetrically partitioned block in
the horizontal direction has a longer length or shorter length.
[94] Then, in determining the length of the filter tap (S630), the length of the filter
tap for interpolation of the PU corresponding to the encoding(or decoding) target block i5 s
determined based on a result of determining the length. As described above, the length
of the filter tap is determined based on a partitioned length in the asymmetric direction.
For instance, the length of the filter tap may be determined such that tap with a shorter
length in the vertical direction than in the horizontal direction is applied to an
10 asymmetrically partitioned block having a shorter length in the vertical direction while tap
with a shorter length in the horizontal direction than in the vertical direction is applied to
an asymmetrically partitioned block having a shorter length in the horizontal direction.
[95] In performing interpolation (S640), the encoding apparatus (and/or decoding
apparatus) performs interpolation based on the length of the filter tap determined in
15 determining the length of the filter (S630).
[96] According to the present embodiment, in the encoding process, interpolation
is performed based on the determined length of the filter tap, and a bitstream is generated
including the information on the length of the filter tap.
[97] FIG. 7 schematically illustrates that an interpolation filter tap adequate for a
20 vertical or horizontal length of a partitioned PU is used when AMP is used in the vertical
direction in the video encoding apparatus according to an exemplary embodiment of the
present invention.
[98] Referring to FIG. 7, interpolation may be performed for an asymmetrically
partitioned block with a larger area using a filter with a longer tap than for an
25 asymmetrically partitioned block with a smaller area. Further, a total length of filter taps
23
in the asymmetrical direction of at least two partitioned blocks may be greater than a
length of a filter tap in a direction other than the asymmetric direction.
[99] Referring to a left illustration of FIG. 7, a 2N x nU block is asymmetrically
partitioned in the vertical direction, wherein an upper block 710 is a partitioned block with
a shorter length. The upper block 710 of the 2N x nU block has a shorter length in th5 e
vertical direction than a lower block 720, in which a ratio between lengths of the upper
block 710 and the lower block 720 may be 16: 48. In this case, a longer-tap filter may be
used for the lower block 720 with a larger area than for the upper block 710 with a smaller
area. Further, a total length of filter taps in the asymmetric direction, that is, a total
10 length of a vertical filter tap of the upper block 710 and a vertical filter tap of the lower
block 720, may be larger than a length of a horizontal filter tap for the upper block 710
and the lower block720.
[100] For example, a 4-tap filter may be used for the upper block 710 in
interpolation for the vertical direction, while a 6-tap filter may be used for the lower block
15 720 in interpolation for the vertical direction. That is, a total tap length of the 4-tap filter
and the 6-tap filter is 10, which is larger than a horizontal filter tap length of 8.
[101] Referring to a right illustration of FIG. 7, the same manner may be applied
to a 2N x nD block, in which case a 6-tap filter may be used for an upper block 730 in
interpolation for the vertical direction, while a 4-tap filter may be used for a lower block
20 740 in interpolation for the vertical direction.
[102] FIG. 8 schematically illustrates that an interpolation filter tap adequate for a
vertical or horizontal length of a partitioned PU is used when AMP is used in the
horizontal direction in the video encoding apparatus according to an exemplary
embodiment of the present invention.
25 [103] Referring to a left illustration of FIG. 8, an nL x 2N block is asymmetrically
24
partitioned in the horizontal direction, wherein a left block 810 is a partitioned block with
a shorter length. The left block 810 of the nL x 2N block has a shorter length in the
horizontal direction than a right block 820, in which a ratio between lengths of the left
block 810 and the right block 820 may be 16: 48. In this case, a longer-tap filter may be
used for the right block 820 with a larger area than for the left block 810 with a smalle5 r
area. Further, a total length of filter taps in the asymmetric direction, that is, a total
length of a horizontal filter tap of the left block 810 and a horizontal filter tap of the right
block 820, may be larger than a length of a vertical filter tap for the left block 810 and the
right block 820.
10 [104] For example, a 4-tap filter may be used for the left block 810 in
interpolation for the horizontal direction, while a 6-tap filter may be used for the right
block 820 in interpolation for the horizontal direction.
[105] Referring to a right illustration of FIG. 8, the same manner may be applied
to an nR x 2N block, in which case a 6-tap filter may be used for a left block 830 in
15 interpolation for the horizontal direction, while a 4-tap filter may be used for a right block
840 in interpolation for the horizontal direction.
[106] Table 1 illustrates vertical and horizontal interpolation filter tap numbers of
asymmetrical blocks.
[Table 1]
Vertical filter tap Horizontal filter tap
2N x nU (upper block) 4 8
2N x nU (lower block) 6 8
2N x nD (upper block) 6 8
2N x nD (lower block) 4 8
25
nL x 2N (left block) 4 8
nL x 2N (right block) 6 8
nR x 2N (left block) 6 8
nR x 2N (right block) 4 8
[108] FIG. 9 is a flowchart schematically illustrating a video encoding method
according to an exemplary embodiment of the present invention.
[109] Referring to FIG. 9, the encoding apparatus may derive a predicted motion
value of a current inter PU (S910). Motion information on the current PU is not
transmitted as it is but differential value from predicated value obtained from temporall5 y
and spatially neighboring blocks are transmitted so as to enhance compression efficiency.
The encoding apparatus may derive a merge candidate list and an advanced motion vector
prediction (AMVP) candidate list for the current inter PU so as to derive the predicted
motion value.
10 [110] The encoding apparatus may generate a PU using the motion information
(S920). Specifically, interpolation may be performed using a short-tap filter for an
asymmetrically partitioned PU in a direction of a shorter partitioned length. Interpolation
methods for the asymmetrically partitioned PU have been described above, and thus
descriptions thereof are omitted herein.
15 [111] The encoding apparatus may encode the motion information on the current
block (S930). In a merge mode, if a candidate having the same motion information as
the current PU is present among merge candidates, the encoding apparatus indicates that
the current PU is in the merge mode and transmits a flag indicating that the merge mode is
used and an index indicating which candidate is used among the merging candidates. In
20 an AMVP mode, the encoding apparatus determines a candidate minimizing a cost
function among AMVP candidates by comparing motion vector information between the
26
AMVP candidates and the current PU, and performs motion compensation using the
determined candidate and a differential value in motion information between the current
PU and the AMVP candidate to obtain a residual signal.
[112] The encoding apparatus may generate a residual block corresponding to the
current block (S940). As described above, the encoding apparatus may perform inte5 r
prediction and/or intra prediction for the current block, thereby generating a prediction
block corresponding to the current block. Here, the encoding apparatus may generate a
residual signal, that is, the residual block, by obtaining a difference by pixels between a
pixel value of the current block and a pixel value of the prediction block.
10 [113] In FIG. 9, the encoding apparatus may transform the residual signal, that is,
the residual block (S950). The encoding apparatus may perform transcoding on the
residual signal by using a transform kernel, and the transform kernel may have a 2x2, 4x4,
8x8, 16x16, 32x32 or 64x64 size. In one exemplary embodiment, a transform coefficient
C for an nxn block may be calculated as follows.
15 [114] [Equation 3]
[115] C(n,n)=T(n,n) x B(n,n) x T(n,n)T
[116] Here, C(n,n) is an nxn transform coefficient matrix, T(n,n) is an nxn
transform kernel matrix, and B(n,n) is an nxn matrix of a residual block.
[117] When a transform coefficient is generated via transformation, the encoding
20 apparatus may quantize the generated transform coefficient.
[118] The encoding apparatus may determine based on RDO which to transmit
among the residual block and the transform coefficient. When prediction is properly
done, the residual block, that is, the residual signal, may be transmitted as it is, without
transcoding. The encoding apparatus may compare cost functions before/after
25 transcoding and select a method involving minimum cost. Here, the encoding apparatus
27
may transmit information on a type of a signal (residual signal or transform coefficient) to
be transmitted with respect to the current block to the decoding apparatus.
[119] In FIG. 9, the encoding apparatus may scan the transform coefficient (S960).
[120] When scanning is done, the encoding apparatus may entropy-encode the
scanned transform coefficient and side information (for example, information on an inte5 r
prediction mode of the current block) (S970). The encoded information may be formed
into a compressed bitstream and be stored in a medium or transmitted through an NAL.
[121] Although the encoding method is described with a series of stages based on
the flowchart in FIG. 9, the present invention is not limited thereto. Some stages of FIG.
10 9 may be carried out in different order from described above or in parallel. Further,
additional stages may be included between stages in the flowchart, or one or more stages
may be deleted from the flowchart of FIG. 9 within the scope of the present invention.
[122] FIG. 10 is a flowchart schematically illustrating a video decoding method
according to an exemplary embodiment of the present invention.
15 [123] Referring to FIG. 10, the decoding apparatus may entropy-decode a
bitstream received from the encoding apparatus (S1010). For instance, the decoding
apparatus may derive a prediction mode and a residual signal of a current block based on a
variable length coding (VLC) table and/or CABAC. The decoding apparatus may obtain
information on whether a signal received with respect to the current block is the residual
20 signal or a transform coefficient. And the decoding apparatus may obtain the residual
signal or a 1D vector of transform coefficients for the current block. When the received
bitstream includes side information needed for decoding, both the bitstream and the side
information may be entropy-decoded.
[124] In FIG. 10, the decoding apparatus may inverse-scan the entropy-decoded
25 residual signal or transform coefficients to generate a two-dimensional block (S1020).
28
Here, a residual block may be generated in the case of the residual signal, and a twodimensional
block of transform coefficients may be generated in the case of the transform
coefficients. When the transform coefficients are generated by entropy-decoding, the
decoding apparatus may dequantize the generated transform coefficients (S1030).
[125] The decoding apparatus may inverse-transform the dequantized transfor5 m
coefficients, thereby generating a residual block (S1040). Inverse transformation may be
represented by Equation 4.
[126] [Equation 4]
[127] B(n,n)=T(n,n) x C(n,n) x T(n,n)T
10 [128] When the residual block is generated, the decoding apparatus may perform
inter prediction based on the residual block (S1050). The decoding apparatus performs
inter prediction using one of the merge mode and the AMVP mode to obtain motion
information.
[129] The decoding apparatus may generate a PU using the obtained motion
15 information. Interpolation may be performed using a short-tap filter for an
asymmetrically partitioned PU in a direction of a shorter partitioned length. Interpolation
methods for the asymmetrically partitioned PU have been described above, and thus
descriptions thereof are omitted herein.
[130] The decoding apparatus may add signal of the residual block and a signal
20 obtained using previous frame to generate a reconstructed block, thereby reconstructing a
picture (S1070). As described above, the decoding apparatus may perform inter
prediction and may perform intra prediction also for a decoding target block to generate a
prediction block corresponding to the decoding target block. Here, the decoding
apparatus may add a pixel value of the prediction block and a pixel value of the residual
25 block by a pixel, thereby generating the reconstructed block.
29
[131]
[132] Although the decoding method is described with a series of stages based on
the flowchart in FIG. 10, the present invention is not limited thereto. Some stages of FIG.
10 may be carried out in different order from described above or in parallel. Further,
additional stages may be included between stages in the flowchart, or one or more stage5 s
may be deleted from the flowchart of FIG. 10 within the scope of the present invention.
[133]
[134] While methods have been described with a series of stages or blocks based
on the flowcharts in the aforementioned embodiments, the present invention is not limited
10 to the foregoing sequence of the stages. Some stages may be carried out in different
order from described above or at the same time. Also, it will be understood by those
skilled in the art that the stages illustrated in the flowcharts are not exclusive, additional
stages may be included in the flowchart, or one or more stages may be deleted from the
flowcharts without affecting the scope of the present invention.
15 [135] The present invention has been described with reference to the exemplary
embodiments, and the foregoing embodiments include various aspects of examples.
Although all possible combinations may not be mentioned to illustrate various aspects, it
will be appreciated by those skilled in the art that changes, modifications and alternatives
may be made in these exemplary embodiments without departing from the principles and
20 spirit of be the invention, the scope of which is defined in the appended claims and their
equivalents.
[136]
30
I/We claim:
1. A video decoding method, the method comprising:
reconstructing a residual value by entropy-decoding received bitstream and
dequantizing and inverse-transforming residual value information5 ;
generating a final prediction unit by performing inter prediction on a prediction unit
which is partitioned from a coding unit into at least two prediction units by asymmetric
motion partitioning (AMP), the two partitioned prediction units comprising a first
partitioned prediction unit and a second partitioned prediction unit; and
10 reconstructing a picture by adding the final prediction unit to the residual values,
wherein the generating of the final prediction unit comprises performing
interpolation using a filter tap with a variable length based on a horizontal length or a
vertical length of the first partitioned prediction unit so that pixels in the second
partitioned prediction unit are not involved in interpolation.
15
2. The method of claim 1, wherein the generating of the final prediction unit
comprises performing interpolation for the first partitioned prediction unit and performing
interpolation for the second partitioned prediction unit using filter taps with different
lengths based on a horizontal length or a vertical length of the prediction units.
20
3. The method of claim 1, wherein the generating of the final prediction unit
comprises performing horizontal interpolation for the first partitioned prediction unit using
a horizontal filter tap shorter than a vertical filter tap when the first partitioned prediction
unit is asymmetric and short in a horizontal direction.
25
31
4. The method of claim 1, wherein the generating of the final prediction unit
comprises performing vertical interpolation for the first partitioned prediction unit using a
vertical filter tap shorter than a horizontal filter tap when the first partitioned prediction
unit is asymmetric and short in a vertical direction.
5
5. The method of claim 1, wherein the generating of the final prediction unit
comprises performing horizontal interpolation for the first partitioned prediction unit using
a horizontally short filter tap shorter than a filter tap for the second partitioned prediction
unit which is horizontally long when the first partitioned prediction unit is asymmetric and
10 short in a horizontal direction.
6. The method of claim 1, wherein the generating of the final prediction unit
comprises performing vertical interpolation for the first partitioned prediction unit using a
vertically short filter tap shorter than a filter tap for the second partitioned prediction unit
15 which vertically long when the first partitioned prediction unit is asymmetric and short in
a vertical direction.
7. The method of claim 1, wherein a 4-tap filter is used for the first
partitioned prediction unit in vertical interpolation and a 6-tap filter is used for the second
20 partitioned prediction unit in vertical interpolation when a 64 x 64 unit to be predicted is
asymmetrically partitioned in a vertical direction into 2N x nU or 2N x nD prediction units,
N being a natural number, a 2N x nU prediction unit being a partitioned form having an
upper block with a smaller area, and a 2N x nD prediction unit being a partitioned form
having a lower block with a smaller area, and a 4-tap filter is used for the first partitioned
25 prediction unit in horizontal interpolation and a 6-tap filter is used for the second
32
partitioned prediction unit in horizontal interpolation when the 64 x 64 unit to be predicted
is asymmetrically partitioned in a horizontal direction into nL x 2N or nR x 2N prediction
units, N being a natural number, an nL x 2N prediction unit being a partitioned form
having a left block with a smaller area, and an nR x 2N prediction unit being a partitioned
form having a right block with a smaller area5 .
8. The method of claim 1, wherein a total length of filter taps in an
asymmetric direction of the first and the second partitioned prediction units is larger than a
length of a filter tap in a direction other than the asymmetric direction.
10
9. The method of claim 1, wherein the received bitstream comprises
information on a prediction mode and a form of a prediction unit corresponding to a
decoding target block.
15 10. The method of claim 9, wherein the received bitstream further comprises
information on a length of an interpolation filter tap of the prediction unit corresponding
to the decoding target block.
11. The method of claim 1, wherein the generating of the final prediction unit
20 comprises obtaining, from the bitstream, partition information on which direction the
partitioned prediction units are asymmetrical; determining, based on the partition
information, which asymmetric direction the partitioned prediction units have a longer
length; determining a length of a filter tap to be used for interpolation based on a
determination result; and performing interpolation using the determined filter tap.
25
33
12. A video decoding apparatus, the apparatus comprising:
a residual value reconstructing module to reconstruct a residual value by entropydecoding
a received bitstream and dequantizing and inverse-transforming residual value
information;
a final prediction unit generating module to generate a final prediction unit b5 y
performing inter prediction on a prediction unit which is a block partitioned from a coding
unit into at least two prediction units by asymmetric motion partitioning (AMP), the two
partitioned prediction units comprising a first partitioned prediction unit and a second
partitioned prediction unit; and
10 a picture reconstructing module to reconstruct a picture by adding the final
prediction unit and the residual value,
wherein the final prediction unit generating module performs interpolation using a
filter tap with a variable length based on a horizontal length or a vertical length of the first
partitioned prediction unit so that pixels in the second partitioned prediction unit are not
15 involved in interpolation.
13. A video encoding method, the method comprising:
performing inter prediction on a prediction unit obtained by partitioning an input
picture using asymmetric motion partitioning (AMP) to predict and encode the picture, the
20 partitioned prediction unit comprising a first partitioned prediction unit and a second
partitioned prediction unit; and
transforming and quantizing a residual value that is a difference between a
prediction unit generated by the inter prediction and a current prediction unit, and entropyencoding
thereon,
25 wherein the performing of the inter prediction comprises performing interpolation
34
using a filter tap with a variable length based on a horizontal length or a vertical length of
the first partitioned prediction unit so that pixels in the second partitioned prediction unit
are not involved in interpolation.
14. The method of claim 13, wherein the performing of the inter predictio5 n
comprises performing interpolation for the first partitioned prediction unit and performing
interpolation for the second partitioned prediction unit using filter taps with different
lengths based on a horizontal length or a vertical length of the prediction units.
10 15. The method of claim 13, wherein the performing of the inter prediction
comprises performing horizontal interpolation for the first partitioned prediction unit using
a horizontal filter tap shorter than a vertical filter tap when the first partitioned prediction
unit is asymmetric and short in a horizontal direction.
15 16. The method of claim 13, wherein the performing of the inter prediction
comprises performing horizontal interpolation using a horizontally short filter tap shorter
than a filter tap for the second partitioned prediction unit which is horizontally long when
the first partitioned prediction unit is asymmetric and short in a horizontal direction.
20 17. The method of claim 13, wherein a total length of filter taps in an
asymmetric direction of the first and the second partitioned prediction units is larger than a
length of a filter tap in a direction other than the asymmetric direction.
18. The method of claim 13, wherein the performing of the inter prediction
25 comprises acquiring information on which direction the partitioned prediction units are
35
asymmetrical; determining, based on the obtained information, which asymmetric
direction the partitioned prediction units have a longer length; determining a length of a
filter tap to be used for interpolation based on a determination result; and performing
interpolation using the determined filter tap.
5
19. The method of claim 13, wherein transforming and quantizing a residual
value, and entropy-encoding thereon comprises generating a bitstream, and
wherein the bitstream comprises information on a length of an interpolation filter
tap of the prediction unit corresponding to an encoding target block.
10
20. A video encoding apparatus, the appratus comprising:
an inter prediction module to perform inter prediction on a prediction unit obtained
by partitioning an input picture using asymmetric motion partitioning (AMP) to predict
and encode the picture, the partitioned prediction unit comprising a first partitioned
15 prediction unit and a second partitioned prediction unit; and
an entropy encoding module to entropy-encode a residual value which is
transformed and/or quantized, wherein the residual value is a difference between a
prediction unit generated by the inter prediction and a current prediction unit,
wherein the inter prediction module performs interpolation using a filter tap with a
20 variable length based on a horizontal length or a vertical length of the first partitioned
36
prediction unit so that pixels in the second partitioned prediction unit are not involved in
interpolation.