The field of the invention is that of coding and decoding 5 images or sequences of
images, and in particular video streams.
More specifically, the invention relates to the compression of images or sequences of
images using a block representation of the images.
The invention can notably be applied to the image or video coding implemented in the
10 current or future encoders (JPEG, MPEG, H.264, HEVC, etc. and their amendments), and to
the corresponding decoding.
2. Prior art
Digital images and sequences of images take up a lot of space in terms of memory,
15 which requires, when transmitting these images, to compress them in order to avoid
congestion problems on the network used for this transmission.
Many techniques for compressing video data are already known. Among these, the
HEVC compression standard (“High Efficiency Video Coding, Coding Tools and
Specification”, Matthias Wien, Signals and Communication Technology, 2015) proposes to
20 implement a prediction of pixels of a current image in relation to other pixels belonging to the
same image (intra prediction) or to a previous or subsequent image (inter prediction).
More specifically, the intra prediction uses the spatial redundancies within an image.
To do this, the images are split into blocks of pixels. The blocks of pixels are then predicted
using already reconstructed information, corresponding to the previously coded/decoded
25 blocks in the current image according to the scanning order of the blocks in the image.
Furthermore, in a standard manner, the coding of a current block is carried out using
a prediction of the current block, referred to as the predictor block, and a prediction residue
or “residual block”, corresponding to a difference between the current block and the predictor
block. The resulting residual block is then transformed, for example using a DCT (discrete
30 cosine transform) type transform. The coefficients of the transformed residual block are then
quantized, coded by entropy coding and transmitted to the decoder, that can reconstruct the
current block by adding this residual block to the predictor block.
The decoding is done image by image, and for each image, block by block. For each
block, the corresponding elements of the stream are read. The inverse quantization and the
35 inverse transform of the coefficients of the residual block are performed. Then, the block
prediction is calculated to obtain the predictor block, and the current block is reconstructed
by adding the prediction (i.e. the predictor block) to the decoded residual block.
2
In US9253508, a DPCM (Differential Pulse Code Modulation) coding technique for
coding blocks in intra mode is integrated into an HEVC encoder. Such a technique consists
in predicting a set of pixels of an intra block by another set of pixels of the same block that
have been previously reconstructed. In US9253508, a set of pixels of the intra block to be
coded corresponds to a row of the block, or a column, or a 5 row and a column, and the intra
prediction used to predict the set of pixels is one of the directional intra predictions defined in
the HEVC standard.
However, such a technique is not optimal. Indeed, the prediction of a pixel by
previously processed neighbouring pixels is well adapted to code natural type data (photos,
10 videos). However, when the type of content is artificial, for example, content corresponding to
screenshots or synthesis images, the images have strong discontinuities generating highenergy
transitions.
More particularly, synthesis images, for example, are likely to contain areas with a
very small number of pixel values, hereinafter also referred to as levels. For example, some
15 areas can have only 2 levels: one for the background and one for the foreground, such as
black text on a white background.
In the presence of such a transition in an area of the image, the value of a pixel to be
coded is then very far from the value of the neighbouring pixels. A prediction of such a pixel
as described above using previously processed neighbouring pixels can then hardly model
20 such transitions.
There is therefore a need for a new coding and decoding method to improve the
compression of image or video data.
3. Summary of the invention
25 The invention improves the state of the art. For this purpose, it relates to a method for
decoding a coded data stream representative of at least one image that is split into blocks.
Such a decoding method comprises, for at least one block of the image, referred to as the
current block:
- determining a group of pixel values that are constant in the block from previously decoded
30 pixels,
- for each pixel of the block:
 decoding a prediction residue associated with said pixel,
 determining a prediction value associated with the pixel according to a first
prediction mode, according to which the pixel is predicted from at least one other
35 previously decoded pixel, said other previously decoded pixel belonging to said
current block or to a previously decoded block of the image,
3
 decoding from the data stream an item of information indicating whether the pixel
is predicted according to a second prediction mode according to which the pixel is
predicted using a prediction resulting from said group of pixel values that are
constant in the block,
 when the item of information indicates that the pixel 5 is predicted according to the
second prediction mode:
o selecting a value of said group,
o replacing said prediction value associated with the pixel with said selected
value,
10  reconstructing said pixel using the prediction value associated with the pixel and
the prediction residue.
Correlatively, the invention also relates to a method for coding a data stream representative
of at least one image that is split into blocks. Such a coding method comprises, for at least
15 one block of the image, referred to as the current block:
- determining a group of pixel values that are constant in the block from previously decoded
pixels,
- for each pixel of the block:
 determining a prediction value associated with the pixel according to a first
20 prediction mode according to which the pixel is predicted by the prediction value
associated with the pixel determined from at least one other previously decoded
pixel, said other previously decoded pixel belonging to said current block or to a
previously decoded block of the image,
 determining a prediction mode for the pixel from the first prediction mode and a
25 second prediction mode according to which the pixel is predicted using a
prediction resulting from said group of pixel values that are constant in the block,
 coding in the data stream, an item of information indicating whether the pixel is
predicted according to the second prediction mode,
 when the item of information indicates that the pixel is predicted according to the
30 second prediction mode:
o selecting a value of said group,
o replacing said prediction value associated with the pixel with said selected
value,
 calculating a prediction residue associated with said pixel using the prediction
35 value associated with the pixel and the value of said pixel,
 coding the prediction residue associated with said pixel in the data stream,
4
 reconstructing said pixel using the prediction value associated with the pixel and
the decoded prediction residue.
The invention thus improves the compression performance of a coding mode using a local
prediction by neighbouring pixels of a pixel to be coded. Advantageously, 5 a group of pixel
values representative of the values of neighbouring pixels of a block to be coded is
determined. For example, this group comprises a predetermined number of pixel values that
are the most frequent among the neighbouring pixels of the block to be coded. Typically, this
group of values can comprise intensity values of the image layers when the image is
10 represented in layers, for example for synthesis images, or comprising areas with a delimited
foreground and background, such as black text on a white background.
According to a particular embodiment of the invention, the group of values comprises two
values representative of the two most frequent values in the neighbourhood of the block.
When a pixel located in a transition area is detected, its prediction value is changed to one of
15 the values of the group thus determined.
The values of such a group are said to be constant in the current block because they are
determined only once for all the pixels of the current block.
According to a particular embodiment of the invention, a value of the group is selected
20 according to a distance between the prediction value associated with said pixel and
determined according to the first prediction mode in relation to the constant pixel values of
the group.
This particular embodiment of the invention allows the convenient selection of a prediction
value of the group for a pixel located in a transition area and does not require additional
25 information to be transmitted to indicate this selection.
According to another particular embodiment of the invention, the group comprising a first
value and a second value, when a distance between the prediction value associated with
said pixel and the first value is less than a distance between the prediction value associated
30 with said pixel and the second value, the selected value of said group is the first value, and
the selected value of said group is the second value otherwise.
According to another particular embodiment of the invention, the item of information
indicating whether the pixel is predicted according to the second prediction mode is decoded
35 from the data stream or coded in the data stream only when the prediction residue of the
pixel is different from 0.
5
This particular embodiment avoids coding the item of information indicating a prediction
according to the second prediction mode when the prediction residue is different from 0.
Thus, according to this particular embodiment, at the decoder, the first prediction mode is
used by default to predict the current pixel.
This particular embodiment of the invention avoids unnecessary information 5 to be coded by
the encoder. Indeed, at the encoder, when the prediction according to the first prediction
mode results in a zero prediction residue, i.e. an optimal prediction, the item of information
indicating that the second prediction mode is not used for the current pixel is implicit.
Such a particular embodiment of the invention can be implemented at the encoder, by a prior
10 step consisting in calculating the prediction residue from the prediction resulting from the first
prediction mode or by a step consisting in determining whether or not the original value of the
pixel to be coded is far from the prediction value resulting from the first prediction mode.
According to another particular embodiment of the invention, the determination of a group of
15 pixel values that are constant in the block from previously decoded pixels is performed by
calculating a histogram of the values of neighbouring pixels of the current block that have
been previously reconstructed and selecting at least two pixel values representative
respectively of two pixel values that are the most frequent among the neighbouring pixels of
the current block.
20
According to another particular embodiment of the invention, a threshold value is determined
from at least one value of said group of pixel values that are constant in the block from
previously decoded pixels. When determining a prediction mode for the pixel, the second
prediction mode is chosen:
25 - when the original value of said pixel is greater than said threshold value and the threshold
value is greater than the prediction value associated with the pixel determined according to
the first prediction mode, or
- when the original value of said pixel is less than said threshold value and the threshold
value is less than the prediction value associated with the pixel determined according to the
30 first prediction mode.
The invention also relates to a device for decoding a coded data stream representative of at
least one image that is split into blocks. Such a decoding device comprises a processor
configured, for at least one block of the image, referred to as the current block, to:
35 - determine a group of pixel values that are constant in the block from previously decoded
pixels,
- for each pixel of the block:
6
 decode a prediction residue associated with said pixel,
 determine a prediction value associated with the pixel from at least one other
previously decoded pixel, said other previously decoded pixel belonging to said
current block or to a previously decoded block of the image,
 determine from the data stream an item of information 5 indicating whether the pixel
is predicted using a prediction resulting from said group of pixel values that are
constant in the block,
 when the item of information indicates that the pixel is predicted using a prediction
resulting from the group of pixel values that are constant in the block:
10 o select a value of said group,
o replace said prediction value associated with the pixel with said selected
value,
 reconstruct said pixel using the prediction value associated with the pixel and the
prediction residue.
15
According to a particular embodiment of the invention, such a decoding device is comprised
in a terminal.
The invention also relates to a device for coding a data stream representative of at least one
20 image that is split into blocks. Such a coding device comprises a processor configured, for at
least one block of the image, referred to as the current block, to:
- determine a group of pixel values that are constant in the block from previously decoded
pixels,
- for each pixel of the block:
25  determine a prediction value associated with the pixel according to a first
prediction mode, according to which the pixel is predicted from at least one other
previously decoded pixel, said other previously decoded pixel belonging to said
current block or to a previously decoded block of the image,
 determine a prediction mode for the pixel from the first prediction mode and a
30 second prediction mode according to which the pixel is predicted using a
prediction resulting from said group of pixel values that are constant in the block,
 code in the data stream an item of information indicating whether the pixel is
predicted according to the second prediction mode,
 when the item of information indicates that the pixel is predicted according to the
35 second prediction mode:
o select a value of said group,
7
o replace said prediction value associated with the pixel with said selected
value,
 calculate a prediction residue associated with said pixel using the prediction value
associated with the pixel and the value of said pixel,
 code the prediction residue associated with sai 5 d pixel in the data stream,
 reconstruct said pixel using the prediction value associated with the pixel and the
decoded prediction residue.
According to a particular embodiment of the invention, such a coding device is comprised in
a terminal, or a server.
10
The invention also relates to a data stream representative of at least one image that is split
into blocks. Such a data stream method comprises, for at least one block of the image,
referred to as the current block, and for each pixel of the current block:
- an item of information representative of a prediction residue associated with said pixel,
15 - an item of information indicating whether the pixel is predicted using a prediction resulting
from a group of pixel values that are constant in the block, the group of pixel values that are
constant in the block being determined from previously decoded pixels.
The decoding method, respectively the coding method, according to the invention can be
20 implemented in various ways, notably in wired form or in software form. According to a
particular embodiment of the invention, the decoding method, respectively the coding
method, is implemented by a computer program. The invention also relates to a computer
program comprising instructions for implementing the decoding method or the coding method
according to any one of the particular embodiments previously described, when said program
25 is executed by a processor. Such a program can use any programming language. It can be
downloaded from a communication network and/or recorded on a computer-readable
medium.
This program can use any programming language, and can be in the form of source code,
object code, or intermediate code between source code and object code, such as in a
30 partially compiled form, or in any other desirable form.
The invention also relates to a computer-readable storage medium or data medium
comprising instructions of a computer program as mentioned above. The recording media
mentioned above can be any entity or device able to store the program. For example, the
medium can comprise a storage means such as a memory. On the other hand, the recording
35 media can correspond to a transmissible medium such as an electrical or optical signal, that
can be carried via an electrical or optical cable, by radio or by other means. The program
according to the invention can be downloaded in particular on an Internet-type network.
8
Alternatively, the recording media can correspond to an integrated circuit in which the
program is embedded, the circuit being adapted to execute or to be used in the execution of
the method in question.
5 4. List of figures
Other characteristics and advantages of the invention will emerge more clearly upon reading
the following description of a particular embodiment, provided as a simple illustrative nonrestrictive
example, and the annexed drawings, wherein:
[Fig 1] Figure 1 shows steps of the coding method according to a particular embodiment of
10 the invention.
[Fig 2A] Figure 2A illustrates an example of a portion of a coded data stream according to a
particular embodiment of the invention.
[Fig 2B] Figure 2B illustrates an example of a portion of a coded data stream according to
another particular embodiment of the invention.
15 [Fig 3A] Figure 3A illustrates a position example of the neighbouring blocks of a current block
to determine an intra prediction mode according to a particular embodiment of the invention.
[Fig 3B] Figure 3B illustrates a position example of the reference pixels used to predict pixels
of a current block according to a particular embodiment of the invention.
[Fig 4] Figure 4 shows steps of the decoding method according to a particular embodiment of
20 the invention.
[Fig 5] Figure 5 illustrates examples of blocks comprising content such as screens each with
two layers of content, and their respective neighbourhood in the image according to a
particular embodiment of the invention.
[Fig 6] Figure 6 illustrates an example of a 16x16 block comprising content such as screens
25 with two layers of content and a transition map showing the transition states of the pixels for
that block according to a particular embodiment of the invention.
[Fig 7] Figure 7 shows the simplified structure of a coding device adapted to implement the
coding method according to any one of the particular embodiments of the invention.
[Fig 8] Figure 8 shows the simplified structure of a decoding device adapted to implement the
30 decoding method according to any one of the particular embodiments of the invention.
5. Description of an embodiment of the invention
5.1 General principle
The invention improves a coding mode of a block of an image using a local prediction for
35 pixels of the block located on a transition between two very distinct levels of pixel values.
A coding mode of a block to be coded using a local prediction allows the use of reference
pixels belonging to the block to be coded to predict other pixels of the block to be coded. This
9
prediction mode reduces the prediction residue by using pixels of the block that are spatially
very close to the pixel to be coded.
However, this coding mode introduces a relatively large coding residue when the original
pixels are far from their prediction. This is generally the case for content such as screenshots
or synthesis images. In this type of content, a block 5 to be coded can have strong
discontinuities. In this case, reference pixels belonging to a background can be used to
predict pixels of the same block belonging to a foreground, or vice versa. In this case, the
item of information available in the reference pixels is not appropriate for an accurate
prediction. The pixels located at the border between a background area and a foreground
10 area are referred to as transition pixels hereafter.
Advantageously, the invention proposes to derive for a block to be coded an item of
information relating to each layer of the image, for example, an item of information relating to
the foreground and an item of information relating to the background, in the case where only
two layers are considered. Additional layers of content can of course be taken into account,
15 increasing the number of items of information to be derived. For example, the derivation of
such information consists in determining a group of pixel values that are constant in the
block.
According to a particular embodiment of the invention, this information relating to each layer
of the image is derived from a local neighbourhood of the block to be coded.
20 Advantageously, this information is used in conjunction with a mechanism for detecting the
transition pixels in the block to be coded. This reduces the residual energy of such pixels.
Figure 5 illustrates blocks (Bi-bl) comprising content such as screens each with two layers of
content, and their respective neighbourhood (Neigh) in the image. As illustrated in figure 5,
the local neighbourhood of a current block to be coded contains useful information relating to
25 the intensity level of the two layers.
According to the invention, when transition pixels in the block to be coded are detected, the
prediction value for these pixels is corrected using an intensity level of the layer
corresponding to the one to which the pixel is likely to belong.
30 According to a particular embodiment of the invention, in order to have an optimal prediction
for each pixel of the block and a limited rate cost, such a mechanism is limited to the pixels
meeting certain conditions.
According to a local neighbourhood of a pixel to be predicted, three states of the pixel to be
predicted can be defined:
35 - s1: the pixel belongs to a homogeneous region in which the local prediction from the
neighbouring pixels is very efficient, for example, it provides a zero quantized prediction
10
residue. In this case, the pixel is not a transition pixel. According to an embodiment variant,
this state can be implicitly detected at the decoder,
- s2: the pixel belongs to a region in which the local prediction from the neighbouring pixels is
moderately efficient, for example, it provides a low prediction residue. The prediction of the
pixel by the above-mentioned correction mechanism is allowed 5 for this pixel, but the
correction mechanism is not applied if the residual prediction error is not large enough
compared to a threshold value determined according to the intensity levels of the layers. In
this case, an indicator is specifically coded to indicate that the correction mechanism is not
used,
10 - s3: the pixel belongs to a region in which the local prediction from the neighbouring pixels is
not efficient, for example, it provides a large prediction residue. The prediction of the pixel by
the above-mentioned correction mechanism is allowed for that pixel, and an indicator is
specifically coded to indicate that use.
15 Figure 6 shows on the left an example of a 16x16 block with light text on a dark background
and on the right a transition map for this block showing how the states described above can
be assigned to the pixels of the block.
5. 2 embodiments
20 Figure 1 shows steps of the coding method according to a particular embodiment of the
invention. For example, a sequence of images I1, I2, ..., INb is coded in the form of a coded
data stream STR according to a particular embodiment of the invention. For example, such a
coding method is implemented by a coding device as described later in relation to figure 7.
A sequence of images I1, I2, ..., INb, Nb being the number of images of the sequence to be
25 coded, is provided as input of the coding method. The coding method outputs a coded data
stream STR representative of the sequence of images provided as input.
In a known manner, the coding of the sequence of images I1, I2, ..., INb is done image by
image according to a coding order previously established and known to the encoder. For
example, the images can be coded in the temporal order I1, I2, ..., INb or in another order, for
30 example I1,I3, I2, ..., INb.
In a step E0, an image Ij to be coded of the sequence of images I1,I2, ..., INb is split into
blocks, for example into blocks of size 32x32 or 64x64 pixels or more. Such a block can be
subdivided into square or rectangular sub-blocks, for example 16x16, 8x8, 4x4, 16x8, 8x16...
In a step E1, a first block or sub-block Xb to be coded of the image Ij is selected according to
35 a predetermined scanning order of the image Ij. For example, it can be the first block in the
lexicographical scanning order of the image.
In a step E2, the encoder chooses the coding mode to code the current block Xb.
11
According to the particular embodiment described here, the encoder selects the coding mode
to code the current block Xb from a first coding mode M1 and a second coding mode M2.
Additional coding modes (not described here) can be used.
According to the particular embodiment described here, the first coding mode M1
corresponds to the conventional intra prediction coding of the 5 current block, for example, as
defined according to the HEVC standard, and the second coding mode M2 corresponds to
an In-Loop Residual (ILR) or DPCM prediction coding described later.
The principle of the invention can be extended to other types of coding modes for the first
coding mode M1. For example, the first coding mode can correspond to any type of coding
10 modes using a transform of the prediction residue (inter-image prediction coding, spatial
prediction with template matching coding, etc.).
In step E2, the encoder can perform a rate/distortion optimisation to determine the best
coding mode to code the current block. During this rate/distortion optimisation, additional
coding modes distinct from the first and the second coding modes can be tested, for example
15 an inter mode coding mode. During this rate/distortion optimisation, the encoder simulates
the coding of the current block Xb according to the different available coding modes in order
to determine the rate and the distortion associated with each coding mode and selects the
coding mode offering the best rate/distortion compromise, for example according to the 𝐷 +
𝜆 × 𝑅 function, where R is the rate required to code the current block according to the
20 evaluated coding mode, D is the distortion measured between the decoded block and the
original current block, and 𝜆 is a Lagrangian multiplier, for example entered by the user or
defined at the encoder.
In a step E20, an item of information indicating the coding mode selected for the current
block is coded in the data stream STR.
25 If the current block Xb is coded according to the first coding mode M1, the method proceeds
to step E21 for coding the block according to M1. If the current block Xb is coded according
to the second coding mode M2, the method proceeds to step E22 for coding the block
according to M2.
Step E21 for coding the block according to the first coding mode M1, according to a
30 particular embodiment of the invention, is described below. According to the particular mode
described here, the first coding mode corresponds to a conventional intra prediction, such as
the one defined in the HEVC standard.
In a step E210, a quantization step 𝛿1 is determined. For example, the quantization step
𝛿1 can be set by the user, or calculated using a quantization parameter setting a compromise
35 between compression and quality and entered by the user or defined by the encoder. Thus,
such a quantization parameter can be the parameter 𝜆 , used in the rate-distortion cost
function 𝐷 + 𝜆 × 𝑅, where D represents the distortion introduced by the coding and R the rate
12
used for coding. This function is used to make coding choices. Typically, a way of coding the
image that minimises this function is sought.
As a variant, the quantization parameter can be QP, corresponding to the quantization
parameter conventionally used in the AVC or HEVC standards. Thus, in the HEVC standard,
the quantization step 𝛿1 is determined by the equation 𝛿5 1 = 𝑙𝑒𝑣𝑒𝑙𝑆𝑐𝑎𝑙𝑒[𝑄𝑃%6] << (𝑄𝑃/6)
where levelScale[ k ] = { 40, 45, 51, 57, 64, 72 } for k = 0..5.
In a step E211, a prediction of the current block is determined using a conventional intra
prediction mode. According to this conventional intra prediction, each predicted pixel is
calculated only from the decoded pixels originating from the neighbouring blocks (reference
10 pixels) located above the current block, and to the left of the current block. The way the
pixels are predicted from the reference pixels depends on a prediction mode that is
transmitted to the decoder, and that is chosen by the encoder from a predetermined set of
modes known to the encoder and the decoder.
Thus, in HEVC there are 35 possible prediction modes: 33 modes that interpolate the
15 reference pixels in 33 different angular directions, and 2 other modes: the DC mode in which
each pixel of the predicted block is produced from the average of the reference pixels, and
the PLANAR mode, that performs a planar and non-directional interpolation. This
“conventional intra prediction” is well known and also used in the ITU-T H.264 standard
(where there are only 9 different modes) as well as in the experimental JEM software
20 available at the Internet address (https://jvet.hhi.fraunhofer.de/), where there are 67 different
prediction modes. In all cases, the conventional intra prediction respects the two aspects
mentioned above (prediction of the pixels of the block to be coded from pixels of the
neighbouring blocks and transmission to the decoder of an optimal prediction mode).
In step E211, the encoder thus chooses one of the available prediction modes from the
25 predetermined list of prediction modes. One way to choose consists for example in
evaluating all the prediction modes and keeping the prediction mode that minimises a cost
function such as, classically, the rate-distortion cost.
In a step E212, the prediction mode chosen for the current block is coded from the
neighbouring blocks of the current block. Figure 3A illustrates a position example of the
30 neighbouring blocks Ab and Bb of the current block Xb to code the prediction mode of the
current block Xb.
In step E212, the intra prediction mode chosen for the current block is coded using the intra
prediction modes associated with the neighbouring blocks.
For example, the approach described in the HEVC standard for coding the prediction mode
35 of the current block can be used. In the example in figure 3A, such an approach consists in
identifying the intra prediction mode mA associated with the block Ab located above the
current block, and the intra prediction mode mB associated with the block Bb located just to
13
the left of the current block. Depending on the value of mA and mB, a list called MPM (for
Most Probable Mode), containing 3 intra prediction modes, and a list called non-MPM,
containing the 32 other prediction modes, are created.
According to the HEVC standard, in order to code the intra prediction 5 mode of the current
block, syntax elements are transmitted:
- a binary indicator indicating if the prediction mode to be coded for the current block is in the
MPM list or not,
- if the prediction mode of the current block belongs to the MPM list, an index in the MPM list
10 corresponding to the prediction mode of the current block is coded,
- if the prediction mode of the current block does not belong to the MPM list, an index in the
non-MPM list corresponding to the prediction mode of the current block is coded.
In a step E213, the prediction residue R for the current block is constructed.
In step E213, in a standard manner, a predicted block P is constructed according to the
15 prediction mode chosen in step E211. Then, the prediction residue R is obtained by
calculating the difference for each pixel between the predicted block P and the original
current block.
In a step E214, the prediction residue R is transformed into RT.
In step E214, a frequency transform is applied to the residue block R in order to produce the
20 block RT comprising transform coefficients. The transform could be a DCT-type transform for
example. It is possible to choose the transform to be used from a predetermined set of
transforms ET and to inform the decoder of the transform used.
In a step E215, the transformed residue block RT is quantized using for example a
quantization step scalar quantization𝛿1. This produces the quantized transformed prediction
25 residue block RTQ.
In a step E216, the coefficients of the quantized block RTQ are coded by an entropy encoder.
For example, the entropy coding specified in the HEVC standard can be used.
In a known manner, the current block is decoded by dequantizing the coefficients of the
quantized block RTQ, then applying the inverse transform to the dequantized coefficients to
30 obtain the decoded prediction residue. The prediction is then added to the decoded
prediction residue in order to reconstruct the current block and obtain its decoded version.
The decoded version of the current block can then be used later to spatially predict other
neighbouring blocks of the image or to predict blocks of other images by inter-image
prediction.
35
Step E22 for coding the block according to the second coding mode M2, according to a
particular embodiment of the invention, is described below. According to the particular
14
embodiment described here, the second coding mode corresponds to an ILR prediction
coding.
In a prior step E220, a quantization step 𝛿2 is determined. For example, the quantization
step 𝛿2 depends on the same quantization parameter as the quantization step 𝛿1 that would
be determined in step E210 if the current block was coded according 5 to the first coding
mode.
According to the invention, in this coding mode, the pixels of the current block can be
predicted according to a first prediction mode or a second prediction mode.
10 According to the first prediction mode, a pixel of the current block is predicted by previously
reconstructed pixels of a neighbouring block of the current block and/or previously processed
pixels of the current block itself. Preferably, to predict a pixel, pixels that are as close as
possible to the pixel to be predicted are chosen. This is why it is referred to a local predictor.
According to the second prediction mode, a pixel of the current block is predicted by a level
15 value of layers selected by a group of values determined, for example, from the
neighbourhood of the current block.
In a step E221, a group of pixel values that are constant in the block is determined from
previously decoded pixels. Several reconstruction levels of the current block are determined,
20 for example two, called f and b. These levels are constructed by analysing values taken by
the reference pixels of the current block, i.e. the pixels from previously processed blocks
neighbouring the current block. There are several techniques for determining the levels f and
b. Thus, it is possible to calculate the histogram of the values of the reference pixels and to
assign b the most frequent value and f the second most frequent value. Another approach
25 consists in identifying the local maxima of the histogram, i.e. the largest values surrounded
by smaller values. The level f is then assigned the largest local maximum and the level b the
second largest local maximum.
Furthermore, according to a particular embodiment of the invention, a threshold value thr is
determined, which is typically halfway between f and b and defined as 𝑡ℎ𝑟 =
(𝑓+𝑏)
2
. In an
alternative embodiment, 𝑡ℎ𝑟 =
𝑑𝑦𝑛
2
30 can also be chosen, where dyn is the maximum value of
the signal.
The embodiment variants described above for determining the group of pixel values that are
constant in the block allow an implicit detection of the image layers which can also be
implemented at the decoder, without the need to transmit additional information.
15
For example, in order to limit the complexity of the detection of the image layers, the direct
neighbourhood of the current block is used: for example, only the pixels of the column on the
left, and of the line above the current block, are used.
According to other variants, more than two values can be determined, by considering the
following local maxima of the 5 histogram, for example.
The values f and b thus determined correspond to the values of the group of values used for
the second prediction mode.
10 The following steps are carried out for each pixel of the current block.
In a step E2201, a local predictor PL for the pixel considered is determined. This local
predictor PL corresponds to the predictor obtained according to the first prediction mode.
The local predictor PL can be determined as follows. If we call X a current pixel to be
predicted of the current block, A the pixel located immediately to the left of X, B the pixel
15 located immediately to the left of and above X, C the pixel located immediately above X, as
illustrated in figure 3B showing a current block Xb. The local predictor PL is defined by:
PL(X) = min(A,B) if C ≥ max(A,B)
max(A,B) if C ≤ min(A,B)
A+B-C otherwise
20 where min(A,B) corresponds to the function returning the smallest value between the value
of A and the value of B and max(A,B) corresponds to the function returning the largest value
between the value of A and the value of B.
Other local prediction functions can be used. According to another variant, several local
prediction functions can be available and the same local prediction function is selected for all
25 the pixels of the current block. For example, an orientation of the texture of the pixels of
previously coded neighbouring blocks is analysed. For example, the previously coded pixels
in a neighbouring block that are located above or to the left of the current block are analysed
using a Sobel-type operator. If it is determined that:
- if no orientation emerges, the prediction function is the one defined above,
30 - if the orientation is horizontal, the prediction function is PL(X)=A,
- if the orientation is vertical, the prediction function is PL(X)=B,
- if the orientation is diagonal, the prediction function is PL(X)=C.
The prediction value PL(X) associated with the current pixel X of the current block is thus
35 obtained according to the location of the pixel in the current block using either pixels outside
the block that are already reconstructed (and thus available with their decoded value), or
pixels previously reconstructed in the current block, or both. In all cases, the predictor PL
16
uses previously reconstructed pixels. In figure 3B, it can be seen that the pixels of the current
block located on the first row and/or the first column of the current block will use as reference
pixels (to construct the prediction value PL(X)) pixels outside the block that are already
reconstructed (pixels in grey in figure 3B) and possibly already reconstructed pixels of the
current block. For the other pixels of the current block, the reference 5 pixels used to construct
the prediction value PL(X) are located inside the current block.
In a step E2202, the prediction mode is determined from the first prediction mode and the
second prediction mode to be used to predict the current pixel.
10 According to a particular embodiment of the invention, the second prediction mode is chosen
when PL(X) < thr < X or when PL(X) > thr > X. In other words, the second prediction mode is
chosen:
- when the original value X of the pixel is greater than the threshold value thr and the
threshold value thr is greater than the prediction value PL(X) associated with the pixel
15 determined according to the first prediction mode, or
- when the original value X of the pixel is less than the threshold value thr and the threshold
value thr is less than the prediction value PL(X) associated with the pixel determined
according to the first prediction mode.
If one of the above conditions is met, then the state of the pixel to be predicted is s=3 and the
20 encoder proceeds to the next step E2203.
In step E2203, an indicator t indicating that the pixel to be predicted is predicted according to
the second prediction mode is set to 1, for example, and encoded in the data stream STR,
for example, by entropy encoding, or transmitted as is in the stream.
In a step E2204, a value of the group of values determined in step E221 is selected to predict
25 the current pixel.
According to a particular embodiment of the invention, a value of the group is selected
according to the distance between the prediction value associated with said pixel determined
according to the first prediction mode in relation to the group pixel values determined in step
E221. For example, when the distance between the prediction value PL(X) associated with
30 said pixel according to the first prediction mode and the value b of the group is less than the
distance between the prediction value PL(X) associated with said pixel according to the first
prediction mode and the value f, the selected value is b, and the selected value is f
otherwise.
The L1 or L2 standard can be used, for example, as a distance measurement.
35 Thus, if |𝑃𝐿(𝑋) − 𝑏| < |𝑃𝐿(𝑋) − 𝑓|, then 𝑃𝐿(𝑋) = 𝑏, otherwise 𝑃𝐿(𝑋) = 𝑓.
The method then proceeds to step E2205.
17
If in step E2202, it is determined that the current pixel is not predicted according to the
second prediction mode, the current pixel is then predicted according to the first prediction
mode. The prediction value PL(X) associated with the current pixel and obtained according to
the first prediction mode is then not modified. The current pixel is then in the state s=1 or
5 s=2.
In a step E2205, a prediction residue R1(X) is calculated for the current pixel as the
difference between the original value X of the current pixel and the prediction value PL(X)
associated with the current pixel, i.e. R1(X)=X-PL(X). Here, the prediction value PL(X) may
10 have been obtained either by the first prediction mode or by the second prediction mode.
The prediction residue R1(X) is then quantized in Q(X), by a quantization step scalar
quantizer 𝛿2 , by 𝑄(𝑋) = 𝑆𝑐𝑎𝑙𝑎𝑟𝑄𝑢𝑎𝑛𝑡(𝑅1(𝑋)) = 𝑆𝑐𝑎𝑙𝑎𝑟𝑄𝑢𝑎𝑛𝑡(𝛿2, 𝑋 − 𝑃𝐿(𝑋)) , the scalar
quantize being for example a nearest-neighbour scalar quantizer such as:
𝑆𝑐𝑎𝑙𝑎𝑟𝑄𝑢𝑎𝑛𝑡(Δ, 𝑥) = 𝑓𝑙𝑜𝑜𝑟 (
𝑥+
Δ
2
Δ
) where Δ is the quantization step and 𝑥 the value to be
15 quantized.
Q(X) is the quantized residue associated with X. It is calculated in the spatial domain, i.e.
calculated directly from the difference between the prediction value PL(X) of the pixel X and
the original value of X. Such a quantized residue Q(X) for the pixel X is stored in a quantized
prediction residue block R1Q, that will be coded later.
20 In a step E2206, the decoded predicted value P1(X) of X is calculated by adding the
dequantized value of the quantized residue Q(X) to the prediction value PL(X). The decoded
predicted value P1(X) of X is thus obtained by 𝑃1(𝑋) = 𝑃𝐿(𝑋) + 𝑆𝑐𝑎𝑙𝑎𝑟𝐷𝑒𝑞𝑢𝑎𝑛𝑡(𝛿2, 𝑄(𝑋)).
For example, the nearest scalar quantization inverse function is given by:
𝑆𝑐𝑎𝑙𝑎𝑟𝐷𝑒𝑞𝑢𝑎𝑛𝑡(Δ, 𝑥) = Δ × 𝑥.
25 The decoded predicted value P1(X) thus makes it possible to predict possible pixels that
remain to be processed in the current block.
Furthermore, the block P1 comprising the decoded/reconstructed values P1(X) of the pixels
of the current block can be defined. Such a block P1 is the ILR predictor of the current block
(as opposed to the conventional intra predictor).
30
According to a particular embodiment of the invention, in a step E2207, when the quantized
prediction residue Q1(X) is not zero, in other words when the amplitude a of the quantized
prediction residue Q1(X) is not zero, the indicator t is set to 0, for example, and coded in the
data stream STR. In this case, the current pixel is considered to be in the state s=2.
35 According to this particular embodiment of the invention, when the quantized prediction
residue Q1(X) is zero, i.e. the amplitude a of the quantized prediction residue Q1(X) is zero,
18
the indicator t is also set to 0 since the current pixel is not predicted by the second prediction
mode, but the indicator t is not coded in the data stream STR. This prediction mode will be
deduced implicitly at the decoder from the decoded value of the amplitude of the quantized
prediction residue Q1(X). In this case, the current pixel is considered to be in the state s=1.
In this case, the method proceeds from step E2206 5 to step E223 directly.
Of course, in practice, when it is explicitly coded (s=2 or s=3), the indicator t is coded in the
data stream after the quantized prediction residue Q1(X) is coded.
According to another particular embodiment of the invention, the indicator t is set to 0, and
10 systematically coded in step E2207 for each pixel, in the data stream STR, regardless of the
value of the amplitude a of the prediction residue Q1(X). Thus, at the decoder, it is explicitly
determined by the decoder whether or not the current pixel is predicted according to the
second prediction mode, by reading the indicator t, regardless of the value of the quantized
prediction residue. In this case, it is not differentiated whether the pixel is in the state s=1 or
15 s=2.
According to a variant, in this particular embodiment of the invention, since the indicator t is
systematically coded, in step E2202, the determination of the prediction mode from the first
prediction mode and the second prediction mode to be used to predict the current pixel can
20 for example be done by comparing a distance measurement between the prediction value
provided by the first prediction mode and the original value X of the current pixel and a
distance measurement between the prediction value provided by the second prediction mode
and the original value X of the current pixel.
25 The steps described above are performed for all the pixels of the current block, in a scanning
order that ensures that the pixels used for the local prediction are available.
According to an embodiment variant, the scanning order of the current block is the
lexicographical order, i.e. from left to right, and from top to bottom.
30 According to another embodiment variant, several scanning orders of the current block can
be used, for example:
- the lexicographical order, or
- scanning the first column from top to bottom, then the column just to the right of it, etc., or
- scanning the diagonals one after the other.
35 According to this other variant, it is possible to simulate the coding cost associated with each
of the scanning orders and to choose the best scanning order for the current block in terms
19
of rate/distortion, then to code for the current block an item of information representative of
the chosen scanning order.
At the end of step E2205, the quantized residue block R1Q was determined. This quantized
residue block R1Q must be coded for transmission to the decoder. 5 The predictor P1 of the
current block was also determined.
In a step E223, the quantized residue block R1Q is coded for transmission to the decoder.
Any known approach such as the method described in HEVC can be used to code the
10 quantized coefficients of a conventional prediction residue.
In a standard manner, each quantized prediction residue Q1(X) of the current block is broken
down into an amplitude value a and a sign indicator sgn when the amplitude a is distinct from
0.
According to the particular embodiment of the invention described here, the amplitude and
15 sign values of the quantized residue block R1Q are coded using an entropy encoder in the
data stream STR.
According to a particular embodiment of the invention, it is possible to determine and code
an additional prediction residue R2 from the ILR predictor obtained for the current block. The
20 coding of an additional prediction residue R2 is, however, optional. It is indeed possible to
simply code the current block by its predicted version P1 and the quantized residue R1Q.
In order to code an additional prediction residue R2 for the current block, the following steps
are implemented.
In a step E224, the difference R2 between the predictor P1 and the original current block Xb
25 is calculated to form an additional residue R2: R2= Xb-P1. The following steps correspond to
the conventional coding steps for this residue R2.
In a step E225, the residue R2 is transformed using a frequency transform in order to
produce the block of coefficients R2T.
The transform can be a DCT-type transform for example. It is possible to choose the
30 transform to be used from a predetermined set of transforms ET2 and to inform the decoder
of the transform used. In this case, the set ET2 can be different from the set ET, in order to
adapt to the particular statistics of the residue R2.
In a step E226, the block of coefficients R2T is quantized, for example using a quantization
step scalar quantization 𝛿. This produces the block R2TQ.
35 The quantization step 𝛿 can be set by the user. It can also be calculated using the parameter
𝜆 setting the compromise between compression and quality and entered by the user or the
20
encoder. For example, the quantization step 𝛿 can correspond to the quantization step 𝛿1 or
be determined similarly to it.
In a step E227, the coefficients of the quantized block R2TQ are then transmitted in a coded
manner. For example, the coding specified in the HEVC standard can be used.
In a known manner, the current block is decoded by dequantizing 5 the coefficients of the
quantized block R2TQ, then applying the inverse transform to the dequantized coefficients to
obtain the decoded prediction residue. The prediction P1 is then added to the decoded
prediction residue in order to reconstruct the current block and obtain its decoded version
Xrec. The decoded version Xrec of the current block can then be used later to spatially predict
10 other neighbouring blocks of the image or to predict blocks of other images by inter-image
prediction.
In a step E23, it is checked if the current block is the last block of the image to be processed
by the coding method, taking into account the previously defined scanning order. If the
15 current block is not the last block of the image to be processed, in a step E24, the
subsequent block of the image to be processed is selected according to the previously
defined scanning order of the image and the coding method proceeds to step E2, where the
selected block becomes the current block to be processed.
If all the blocks of the image have been coded, the method proceeds to the application of the
20 post-processing methods to be applied to the reconstructed image in a step E231. For
example, such post-processing methods can be a deblocking filtering and/or an SAO
(Sample Adaptive Offset) method as defined in the HEVC standard.
The method proceeds to coding (step E25) the next image of the video, if any.
25 Figures 2A and 2B schematically illustrate a portion of a data stream resulting from the
coding as described above according to different particular embodiments of the invention.
Figure 2A illustrates an example of a stream for three pixels (X1, X2, X3) of a block of the
image coded according to a particular embodiment of the invention, wherein it has been
determined that the pixel X1 is considered to be in the state s=3, the pixel X2 is considered
30 to be in the state s=2 and the pixel X3 is considered to be in the state s=1.
It can be seen that according to the variant described here, the data coded for the pixel X1 is
the amplitude value of the quantized prediction residue a(X1), its sign sgn(X1) and the value
of the indicator t set to 1. The data coded for the pixel X2 is the amplitude value of the
quantized prediction residue a(X2), its sign sgn(X2) and the value of the indicator t. For X2,
35 the amplitude value of the quantized prediction residue being distinct from 0, the indicator t
set to 0 is explicitly coded in the stream.
21
The data coded for the pixel X3 is the amplitude value of the quantized prediction residue
a(X3) that is zero. In this case, the amplitude value of the quantized prediction residue is
distinct from 0, so the indicator t set to 0 is not explicitly coded in the stream and will be
implicitly deduced at the decoder.
5
Figure 2B illustrates an example of a stream for three pixels (X1, X2, X3) of a block of the
image coded according to another particular embodiment of the invention, wherein it has
been determined that the pixel X1 is considered to be in the state s=3, the pixel X2 is
considered to be in the state s=2 and the pixel X3 is considered to be in the state s=1.
10 It can be seen that according to the variant described here, the data coded for the pixel X1 is
the amplitude value of the quantized prediction residue a(X1), its sign sgn(X1) and the value
of the indicator t set to 1. The data coded for the pixel X2 is the amplitude value of the
quantized prediction residue a(X2), its sign sgn(X2) and the value of the indicator t set to 0.
The data coded for the pixel X3 is the amplitude value of the quantized prediction residue
15 a(X3) that is zero and the indicator t set to 0.
Figure 4 shows steps of the method for decoding a stream STR of coded data representative
of a sequence of images I1, I2, ..., INb to be decoded according to a particular embodiment of
the invention.
20 For example, the data stream STR was generated via the coding method shown in relation to
figure 1. The data stream STR is provided as input to a decoding device DEC, as described
in relation to figure 8.
The decoding method decodes the stream image by image and each image is decoded block
by block.
25 In a step E40, an image Ij to be decoded is subdivided into blocks. Each block will undergo a
decoding operation consisting in a series of steps that are detailed hereafter. Blocks can be
the same size or different sizes.
In a step E41, a first block or sub-block Xb to be decoded of the image Ij is selected as the
current block according to a predetermined scanning order of the image Ij. For example, it
30 can be the first block in the lexicographical scanning order of the image.
In a step E42, an item of information indicating a coding mode for the current block is read
from the data stream STR. According to the particular embodiment described here, this item
of information indicates if the current block is coded according to a first coding mode M1 or
according to a second coding mode M2. According to the particular embodiment described
35 here, the first coding mode M1 corresponds to the conventional intra prediction coding of the
current block, for example as defined according to the HEVC standard, and the second
coding mode M2 corresponds to the In-Loop Residual (ILR) prediction coding.
22
In other particular embodiments, the item of information read from the stream STR can also
indicate the use of other coding modes to code the current block (not described here).
The step E43 for decoding the current block when the current block is coded according to the
first coding mode M1 is described below.
In a step E430, a quantization step 𝛿1 is determined. For example, 5 the quantization step 𝛿1 is
determined from a quantization parameter QP transmitted in the data stream STR or similarly
to what was done at the encoder. For example, the quantization parameter QP can be the
quantization parameter conventionally used in the AVC or HEVC standards. Thus, in the
HEVC standard, the quantization step 𝛿1 is determined by the equation 𝛿1 =
10 𝑙𝑒𝑣𝑒𝑙𝑆𝑐𝑎𝑙𝑒[𝑄𝑃%6] << (𝑄𝑃/6), where levelScale[ k ] = { 40, 45, 51, 57, 64, 72 } for k = 0..5.
In a step E431, the prediction mode chosen to code the current block is decoded from the
neighbouring blocks. For this purpose, as it was done at the encoder, the intra prediction
mode chosen for the current block is coded using the intra prediction modes associated with
15 the neighbouring blocks of the current block.
The construction of both MPM and non-MPM lists is strictly similar to what was done during
coding. According to the HEVC standard, syntax elements of the following type are decoded:
- a binary indicator indicating if the prediction mode to be coded for the current block is in the
20 MPM list or not,
- if the prediction mode of the current block belongs to the MPM list, an index in the MPM list
corresponding to the prediction mode of the current block is read,
- if the prediction mode of the current block does not belong to the MPM list, an index in the
non-MPM list corresponding to the prediction mode of the current block is read.
25 The binary indicator and the prediction mode index are thus read for the current block from
the data stream STR, to decode the intra prediction mode of the current block.
In a step E432, the decoder constructs a predicted block P for the current block from the
decoded prediction mode.
In a step E433, the decoder decodes the coefficients of the quantized block RTQ from the
30 data stream STR, for example using the decoding specified in the HEVC standard.
In a step E434, the decoded block RTQ is dequantized, for example using a quantization step
scalar dequantization 𝛿1. This produces the block of dequantized coefficients RTQD.
In a step E435, an inverse frequency transform is applied to the block of dequantized
coefficients RTQD in order to produce the decoded prediction residue block RTQDI. The
35 transform could be an inverse DCT-type transform for example. It is possible to choose the
transform to be used from a predetermined set of transforms ETI by decoding an indicator
from the data stream STR.
23
In a step E436, the current block is reconstructed from the predicted block P obtained in step
E432 and the decoded residue block RTQDI obtained in step E435, in order to produce the
decoded current block Xrec, by Xrec = P + RTQDI.
The step E44 for decoding the current block when the current block 5 is coded according to the
second coding mode M2 is described below.
In a step E440, the quantization step 𝛿2 is determined, similarly to what was done at the
encoder.
10 According to the invention, in this coding mode M2, the pixels of the current block can be
predicted according to the first prediction mode or the second prediction mode already
presented in relation to figure 1.
In a step E441, the group of pixel values that are constant in the block is determined from
previously decoded pixels of the image, similarly to what was done at the encoder. It is
15 considered as with the encoder that the level values f and b have been determined.
The following steps are carried out for each pixel of the current block.
In a step E4411, the prediction value of the current pixel according to the first prediction
mode is determined. For this purpose, the same local predictor PL as at the encoder is used.
20 When several local predictors are possible, the local predictor PL is determined similarly to
what was done at the encoder.
In a step E442, the quantized residue R1Q is decoded from the data stream STR. Any known
approach such as the method described in HEVC can be used to decode the quantized
25 coefficients of the conventional prediction residue. The amplitude a of the quantized
prediction residual Q1'(X) for the current pixel is then obtained.
According to a particular embodiment of the invention, in a step E4421, when the amplitude a
of the quantized prediction residue Q1'(X) is zero, an indicator t indicating whether the
current pixel is predicted according to the second prediction mode is implicitly set to 0. In this
30 case, the current pixel is considered to be in the state s=1, and it will be predicted by the
prediction value resulting from the first prediction mode. The quantized prediction residue
Q1'(X) is then reconstructed by 𝑄1′(𝑋) = 0.
Otherwise, when the amplitude a of the quantized prediction residue Q1'(X) is not zero, the
35 sign sgn associated with the quantized prediction residue Q1'(X) is read in the data stream
STR. The quantized prediction residue Q1'(X) is then reconstructed by 𝑄1′(𝑋) = 𝑎 × 𝑠𝑔𝑛.
24
Then, in a step E4422, the indicator t for the current pixel is read in the data stream STR. If
the value of the indicator t read is 0, the state of the current pixel is s=2. If the value of the
indicator t read is 1, the state of the current pixel is s=3.
According to another particular embodiment of the invention, the 5 indicator t is systematically
coded for each pixel of the current block. In this case, in step E4422, the value 0 or 1 of the
indicator t is read in the data stream STR and the state of the pixel s is set accordingly.
When the state of the current pixel is s=3, the current pixel is predicted according to the
10 second prediction mode. In this case, in a step E4423, a value of the group of values
determined in step E441 is selected and assigned to the prediction value PL(X) associated
with the current pixel to predict the current pixel similarly to what was done at the encoder.
For example, if |𝑃𝐿(𝑋) − 𝑏| < |𝑃𝐿(𝑋) − 𝑓|, then 𝑃𝐿(𝑋) = 𝑏, otherwise 𝑃𝐿(𝑋) = 𝑓.
The method then proceeds to step E443.
15
When the state of the current pixel is s= 2 or s=1, the current pixel is predicted according to
the first prediction mode. In this case, the prediction value PL(X) of the current pixel
determined according to the first prediction mode in step E4411 is not changed.
20 In a step E443, the quantized residue Q1'(X) is dequantized using the quantization step 𝛿2,
in order to produce the dequantized residue QD1(X).
In a step E444, the reconstructed value of the current pixel X' is obtained using the prediction
value PL(X) determined in step E4411 or E4423 and the dequantized prediction residue
25 QD1(X): X'=PL(X)+QD1(X).
The prediction residues Q1(X) of the pixels of the current block are placed in a prediction
residue block R1Q, the dequantized prediction residues QD1(X) of the pixels of the current
block are placed in a dequantized prediction residue block R1QD, the reconstructed values X'
30 of the pixels of the current block are placed in a reconstructed block P1.
The above steps are implemented for all the pixels of the current block, in a scanning order
that ensures that the pixels used for the local prediction are available.
For example, the scanning order is the lexicographical order (from left to right, then rows
35 from top to bottom).
25
According to a particular embodiment of the invention, the block P1 comprising the
reconstructed values PL(X)+QD1(X) of each pixel of the current block forms here the
decoded current block Xrec.
According to another particular embodiment of the invention, 5 it is considered that an
additional prediction residue was coded for the current block. It is therefore necessary to
decode this additional prediction residue in order to reconstruct the decoded version of the
current block Xrec.
For example, this other particular embodiment can be activated or not by default at the
10 encoder and decoder level. Or, an indicator can be coded in the data stream with the block
level information to indicate for each block coded according to the ILR coding mode if an
additional prediction residue is coded. Or further, an indicator can be coded in the data
stream with the image or sequence of images level information to indicate for all the blocks of
the image or of the sequence of images coded according to the ILR coding mode if an
15 additional prediction residue is coded.
When an additional prediction residue is coded for the current block, in a step E445, the
coefficients of the quantized prediction residue R2TQ are decoded from the data stream STR,
using means adapted to those implemented at the encoder, for example the means
implemented in an HEVC decoder.
20 In a step E446, the block of quantized coefficients R2TQ is dequantized, for example using a
quantization step scalar dequantization 𝛿1 . This produces the block of dequantized
coefficients R2TQD.
In a step E447, an inverse frequency transform is applied to the block R2TQD in order to
produce the decoded prediction residue block R2TQDI.
25 The inverse transform could be an inverse DCT-type transform for example.
It is possible to choose the transform to be used from a predetermined set of transforms ET2
and to decode the item of information informing the decoder of the transform to be used. In
this case, the set ET2 is different from the set ET, in order to adapt to the particular statistics
of the residue R2.
30 In a step E448, the current block is reconstructed by adding the predicted block P1 obtained
in step E444 to the decoded prediction residue R2TQDI.
In a step E45, it is checked if the current block is the last block of the image to be processed
35 by the decoding method, taking into account the previously defined scanning order. If the
current block is not the last block of the image to be processed, in a step E46, the
subsequent block of the image to be processed is selected according to the previously
26
defined scanning order of the image and the decoding method proceeds to step E42, the
selected block becoming the current block to be processed.
If all the blocks of the image have been coded, the method proceeds to the application of the
post-processing methods to be applied to the reconstructed image in a step E451 if required.
Such post-processing methods can be a deblocking 5 filtering and/or an SAO method.
The method then proceeds to decoding (step E47) the next image of the video, if any.
Figure 7 shows the simplified structure of a coding device COD adapted to implement the
10 coding method according to any one of the particular embodiments of the invention.
According to a particular embodiment of the invention, the steps of the coding method are
implemented by computer program instructions. For this purpose, the coding device COD
has the standard architecture of a computer and notably comprises a memory MEM, a
processing unit UT, equipped for example with a processor PROC, and driven by the
15 computer program PG stored in the memory MEM. The computer program PG comprises
instructions for implementing the steps of the coding method as described above, when the
program is executed by the processor PROC.
At initialisation, the code instructions of the computer program PG are for example loaded
into a RAM memory (not shown) before being executed by the processor PROC. In
20 particular, the processor PROC of the processing unit UT implements the steps of the coding
method described above, according to the instructions of the computer program PG.
Figure 8 shows the simplified structure of a decoding device DEC adapted to implement the
decoding method according to any one of the particular embodiments of the invention.
25 According to a particular embodiment of the invention, the decoding device DEC has the
standard architecture of a computer and notably comprises a memory MEM0, a processing
unit UT0, equipped for example with a processor PROC0, and driven by the computer
program PG0 stored in the memory MEM0. The computer program PG0 comprises
instructions for implementing the steps of the decoding method as described above, when
30 the program is executed by the processor PROC0.
At initialisation, the code instructions of the computer program PG0 are for example loaded
into a RAM memory (not shown) before being executed by the processor PROC0. In
particular, the processor PROC0 of the processing unit UT0 implements the steps of the
decoding method described above, according to the instructions of the computer program
35 PG0.

WE CLAIMS

1. Method for decoding a coded data stream representative of at least one image, said image
being split into blocks, the decoding method comprises, for at least one block of the image,
referred 5 to as the current block:
- determining (E441) a group of pixel values that are constant in the block from previously
decoded pixels,
- for each pixel of the block:
(i) decoding (E442) a prediction residue associated with said pixel,
10 (ii) determining (E4411) a prediction value associated with the pixel according to a
first prediction mode, according to which the pixel is predicted from at least one other
previously decoded pixel, said other previously decoded pixel belonging to said current
block,
(iii) decoding (E4422) from the data stream an item of information indicating whether
15 the pixel is predicted according to a second prediction mode according to which the pixel is
predicted using a prediction resulting from said group of pixel values that are constant in the
block,
(iv) when the item of information indicates that the pixel is predicted according to the
second prediction mode:
20
(a) selecting (E4423) a value of said group,
(b) replacing (E4423) said prediction value associated with the pixel with said
selected value,
(v) reconstructing (E444) said pixel using the prediction value associated with the
25 pixel and the prediction residue.
2. Method for coding a data stream representative of at least one image, said image being
split into blocks, the coding method comprises, for at least one block of the image, referred to
as the current block:
30 - determining (E221) a group of pixel values that are constant in the block from previously
decoded pixels,
- for each pixel of the block:
(i) determining (E2201) a prediction value associated with the pixel according to a first
prediction mode, according to which the pixel is predicted from at least one other previously
35 decoded pixel, said other previously decoded pixel belonging to said current block,
28
(ii) determining (E2202) a prediction mode for the pixel from the first prediction mode
and a second prediction mode according to which the pixel is predicted using a prediction
resulting from said group of pixel values that are constant in the block,
(iii) coding (E2203, E2207) in the data stream an item of information indicating
whether the pixel is predicted according to 5 the second prediction mode,
(iv) when the item of information indicates that the pixel is predicted according to the
second prediction mode:
(a) selecting (E2204) a value of said group,
10 (b) replacing (E2204) said prediction value associated with the pixel with said
selected value,
(v) calculating (E2205) a quantized prediction residue associated with said pixel using
the prediction value associated with the pixel and the value of said pixel,
(vi) reconstructing (E2206) said pixel using the prediction value associated with the
15 pixel and the decoded prediction residue,
(vii) coding (E223) the quantized prediction residue associated with said pixel in the
data stream.
3. Decoding method according to claim 1 or coding method according to claim 2, wherein a
20 value of said group to be used is selected according to a distance between the prediction
value associated with said pixel compared to the constant pixel values of the group.
4. Method according to claim 3, wherein the group comprising a first value and a second
value, when a distance between the prediction value associated with said pixel and the first
25 value is less than a distance between the prediction value associated with said pixel and the
second value, the selected value of said group is the first value, and the selected value of
said group is the second value otherwise.
5. Decoding method according to claim 1 or coding method according to claim 2, wherein the
30 decoding from the data stream or the coding in the data stream of the item of information
indicating whether the pixel is predicted according to the second prediction mode is
performed only when the prediction residue is different from 0.
6. Decoding method according to claim 1 or coding method according to claim 2, wherein the
35 determination of a group of pixel values that are constant in the block from previously
decoded pixels is performed by calculating a histogram of the values of neighbouring pixels
of the current block that were previously reconstructed and the selection at least two pixel
29
values representative respectively of two pixel values that are the most frequent among the
neighbouring pixels of the current block.
7. Coding method according to claim 2, wherein a threshold value is determined from at least
one value of said group of pixel values that are constant in the block from previously
decoded pixels, when determining a prediction mode for the 5 pixel, the second prediction
mode is chosen:
- when the original value of said pixel is greater than said threshold value and the
threshold value is greater than the prediction value associated with the pixel determined
according to the first prediction mode, or
10 - when the original value of said pixel is less than said threshold value and the
threshold value is less than the prediction value associated with the pixel determined
according to the first prediction mode.
8. Device for decoding a coded data stream representative of at least one image, said image
15 being split into blocks, the decoding device comprises a processor (PROC0) configured, for
at least one block of the image, referred to as the current block, to:
- determine a group of pixel values that are constant in the block from previously decoded
pixels,
- for each pixel of the block:
20 (i) decode a prediction residue associated with said pixel,
(ii) determine a prediction value associated with the pixel from at least one other
previously decoded pixel, said other previously decoded pixel belonging to said current
block,
(iii) determine from the data stream an item of information indicating whether the pixel
25 is predicted using a prediction resulting from said group of pixel values that are constant in
the block,
(iv) when the item of information indicates that the pixel is predicted using a prediction
resulting from the group of pixel values that are constant in the block:
(a) select a value of said group,
30 (b) replace said prediction value associated with the pixel with said selected
value,
(v) reconstruct said pixel using the prediction value associated with the pixel and the
prediction residue.
35
30
9. Device for coding a data stream representative of at least one image, said image being
split into blocks, the coding device comprises a processor (PROC) configured, for at least
one block of the image, referred to as the current block, to:
- determine a group of pixel values that are constant in the block from previously decoded
5 pixels,
- for each pixel of the block:
(i) determine a prediction value associated with the pixel according to a first prediction
mode, according to which the pixel is predicted from at least one other previously decoded
pixel, said other previously decoded pixel belonging to said current block,
10 (ii) determine a prediction mode for the pixel from the first prediction mode and a second
prediction mode according to which the pixel is predicted using a prediction resulting from
said group of pixel values that are constant in the block,
(iii) code in the data stream an item of information indicating whether the pixel is
predicted according to the second prediction mode,
15 (iv) when the item of information indicates that the pixel is predicted according to the
second prediction mode:
(a) select a value of said group,
(b) replace said prediction value associated with the pixel with said selected value,
20
(v) calculate a prediction residue associated with said pixel using the prediction value
associated with the pixel and the value of said pixel,
(vi) code the prediction residue associated with the pixel in the data stream,
25 (vii) reconstruct said pixel using the prediction value associated with the pixel and the
decoded prediction residue.
10. Data stream representative of at least one image, said image being split into blocks, the
data stream comprises, for at least one block of the image, referred to as the current block,
30 and for each pixel of the current block:
- an item of information representative of a prediction residue associated with said pixel,
- an item of information indicating whether the pixel is predicted using a prediction resulting
from a group of pixel values that are constant in the block, the group of pixel values that are
constant in the block being determined from previously decoded pixels.
35
31
11. Computer program comprising instructions for implementing the decoding method
according to any one of claims 1 or 3 to 6 or the coding method according to any one of
claims 2 to 7, when said program is executed by a processor.
12. Computer-readable data medium, comprising instructions 5 of a computer program
according to claim 11.