FR3050858A1

FR3050858A1 - IMAGE ENCODING AND DECODING METHOD, IMAGE ENCODING AND DECODING DEVICE AND CORRESPONDING COMPUTER PROGRAMS

Info

Publication number: FR3050858A1
Application number: FR1653884A
Authority: FR
Inventors: Pierrick Philippe; Adria Arrufat
Original assignee: Orange SA
Current assignee: Orange SA
Priority date: 2016-04-29
Filing date: 2016-04-29
Publication date: 2017-11-03

Abstract

L'invention concerne le codage d'au moins une image (ICj) découpée en blocs de données, mettant en œuvre, pour un bloc courant (Bi) à coder de ladite image, ledit bloc courant contenant M vecteurs lignes et N vecteurs colonnes de données, tel que M≥2 et N≥2, un traitement des données du bloc courant par application (C4) d'une transformée sélectionnée selon un critère de performance de codage prédéterminé, dans un ensemble (E) comprenant au moins deux transformées, une transformée dudit ensemble mettant en œuvre les opérations suivantes : - premier traitement de données appliqué sur les M ou N vecteurs du bloc courant, à l'issue duquel est obtenu un bloc de données traité, - deuxième traitement de données appliqué sur les N ou M vecteurs du bloc de données traité, à l'issue duquel est obtenu un bloc de données transformé, le procédé de codage étant caractérisé en ce que au moins une autre transformée dudit ensemble met en œuvre : - soit uniquement un traitement des M lignes de données du bloc courant, - soit uniquement un traitement des N colonnes de données du bloc courant.The invention relates to the coding of at least one image (ICj) cut into data blocks, implementing, for a current block (Bi) to be encoded of said image, said current block containing M row vectors and N column vectors of data, such as M≥2 and N≥2, processing the data of the current block by applying (C4) a selected transform according to a predetermined coding performance criterion, in a set (E) comprising at least two transforms, a transform of said set implementing the following operations: first data processing applied to the M or N vectors of the current block, at the end of which a processed data block is obtained; second data processing applied to the N or M vectors of the processed data block, at the end of which is obtained a transformed data block, the coding method being characterized in that at least one other transform of said set implements: - either only a processing of the M data lines of the current block, or only a processing of N data columns of the current block.

Description

METHOD FOR ENCODING AND DECODING IMAGES APPARATUS FOR ENCODING AND DECODING IMAGES AND CORRESPONDING COMPUTER PROGRAMS

Field of the invention

The present invention relates generally to the field of image processing, and more specifically to the encoding and decoding of digital images and digital image sequences.

The encoding / decoding of digital images applies in particular to images originating from at least one video sequence comprising: images coming from the same camera and succeeding one another temporally (coding / decoding of 2D type), images from different cameras oriented according to different views (coding / decoding of 3D type), - corresponding texture and depth components (coding / decoding of 3D type), - etc ...

The present invention applies similarly to the coding / decoding of 2D or 3D type images. The invention may especially, but not exclusively, apply to video coding implemented in current AVC and HEVC video encoders and their extensions (MVC, 3D-AVC, MV-HEVC, 3D-HEVC, etc.), and to corresponding decoding.

Prior art

Current video encoders (MPEG, H.264, HEVC, ...) use a block representation of the video sequence. The images are cut into blocks, which can be recursively redrawn. Then each block is coded by intra-image prediction or inter-image. Thus, some images are coded by spatial prediction (Intra prediction), other images are also coded by temporal prediction (Inter prediction) with respect to one or more coded-decoded reference images, by means of a compensation in movement known by those skilled in the art.

For each block is coded a residue block, also called prediction residue, corresponding to the original block minus a prediction. The residue blocks are transformed using a transform mathematical operation. At the end of such a transformation, blocks of coefficients are obtained, the coefficients of each block being traversed in a determined travel order, and then quantized by means of a mathematical quantization operation, for example of the scalar type. Monodimensional lists of coefficients are obtained after the quantization.

For the sake of simplification, the mathematical operation of transform will be called later 'transformed' and the mathematical operation of quantification will be called later 'quantification'.

The coefficients of this list are then encoded in bits by an entropy coding whose purpose is to code the coefficients without loss.

The bits obtained after entropy coding are written in a data stream signal which is intended to be transmitted to the decoder.

In a manner known per se, such a signal comprises: the quantized coefficients contained in the aforementioned list, information representative of the coding mode used, in particular: the prediction mode (Intra prediction, Inter prediction, default prediction realizing a prediction for which no information is transmitted to the decoder ("in English" skip ")); • information specifying the type of prediction (orientation, reference image, ...); • the type of division of the block; • movement information if necessary; • etc.

Once the stream has been received by the decoder, 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, the reverse course operation, and the inverse transform of the block coefficients are performed to produce the decoded prediction residue. Then, the prediction of the block is calculated and the block is reconstructed by adding the prediction to the decoded prediction residue.

The conventional coding / decoding technique that has just been described certainly allows improvements in coding performance. Depending on the video context, it allows in particular: - an improvement in the quality of the images for a given bit rate of the network used to transmit the images, - a reduction of the transmission rate of the images for a previously fixed image quality criterion.

In the field of video coding, the discrete cosine transforms, DCT (abbreviation of "Discrete Cosine Transform"), or the discrete sinus transforms, DST (abbreviation of "Discrete Sine Transform"), are generally preferred, especially for the following reasons: - they are block transforms and it is thus easy to manipulate the blocks independently of each other, - they are effective for compacting the information in the frequency domain, where the flow reduction operation operates. they have fast implementation methods that require the order M * log2 (M) operations, M being the number of transformed coefficients.

The aforementioned DCT or DST transforms are implemented separably.

A separable transform can be applied in two different cases.

According to a first case, a first transform Ac is applied to a residue block x composed of K pixels which are organized in the form of a matrix MxN, where Ac is a matrix of data of size MxM and Μ , N are natural numbers greater than or equal to 1. After the application of this first transform is obtained a first transformed block Ac.x.

A transposition operation t is then applied to the transformed block Ac.x. At the end of this transposition, a transposed block (Ac.xÿ.

Finally, a second Al transform is applied to the transposed block (Ac.x) 1, where Al is a data matrix of size NxN. At the end of the application of this second transform is obtained a second transformed block X of K = NxM pixels, such that:

In a second case, the order of application of the Al and Ac transforms is reversed. The second transformed block X of K = NxM pixels is then written as follows:

The transformed block X obtained according to this second case is similar to the transformed block X obtained according to the first case, to a close transposition.

In the particular case where the residue block x is square, that is to say M = N, the matrices Al and Ac have the same size.

On decoding, in a manner known per se, reverse transforms of those mentioned above are applied.

Thus, if the transform has been applied according to the first case, the corresponding inverse transform makes it possible to obtain the residue block x using the following calculation:

Thus, if the transform has been applied according to the second case, the corresponding inverse transform makes it possible to obtain the residue block x by means of the following calculation:

AI'1 and Ac'1 represent the respective inverse transforms of the Al and Ac transforms. They make it possible to obtain the values of the residue block x from the values of the transformed block X. The matrices Al'1 and Ac'1 are commonly called inverse matrices of Al and Ac respectively, in the case where the matrices are chosen orthogonal to each other. correspond to the matrices transposed of Al and Ac respectively.

In addition to the separable type transforms, there are also non-separable type transforms, such as, for example, the Karhunen-Loeve transform (KLT) which is considered to provide an optimal decorrelation of the data of a block considered. Non-separable transforms have the advantage of being able to exploit correlations between any (or more) pair of data within a given block, unlike separable-type transforms that can only exploit the correlations between data. sharing either the same line or the same column of a block considered through transforms that Al and Ac that act independently. Such a difference makes the non-separable type transforms more efficient than the separable transforms, in particular from the point of view of the concentration of the energy on few coefficients and the coding performances. On the other hand, non-separable transforms are very complex in, which makes them difficult to implement in current video encoders. This is the reason why the DCT or DST transforms are currently preferred, especially since they may in certain cases be a good approximation of the KLT transform.

A non-separable type transform is mathematically written as the multiplication of the residual block x, set in the form of a vector of dimension 1xK, by a matrix A of size KxK. The transformed block X obtained after the application of this transform is then written as follows:

At decoding, the inverse transform consists of multiplying the transformed block X by the inverse matrix A'1 of A which can be the transpose of A, when A is orthogonal. Such an inverse transform makes it possible to obtain the following residue block x:

In the field of video coding, it has been proposed, in particular in the publication "Rate-distortion optimized transform competition for intra coding in HEVC, Adrià Arrufat et al. VCIP 2014 "to increase the number of transforms to select at the encoding. For this purpose: - for 4x4 sized blocks, sixteen transformed, previously optimized according to a rate-distortion criterion well known to those skilled in the art, are provisioned, - for blocks of size 8x8, thirty-two transformed, previously optimized according to the debit-distortion criterion are provisioned.

For a current block having a given size, the video encoder selects the transform that minimizes the rate-distortion criterion, by putting into competition the typical transforms, whether of the DCT or DST type, and the optimized transforms. The encoder then signals the decoder which transform has been selected. The decoder uses the inverse transform of that applied to the encoder.

In this document, it is proposed to select an optimal transform from two types of transforms: separable type transforms and non-separable type transforms, as described above.

Such a solution certainly makes it possible to improve coding performance. However, the amount of transforms to be added has an impact on the memory resources to be used by both the encoder and also the decoder which must know the transforms to apply the inverse transform to that applied to the encoding.

In addition, the competition of the proposed transforms requires the implementation of a large number of operations, such as multiplications and additions inherent to the matrix products. The complexity of the calculations is also increased because of the presence of non-separable transforms which do not have, at first sight, fast implementation methods.

Object and summary of the invention

One of the aims of the invention is to overcome disadvantages of the state of the art mentioned above. For this purpose, an object of the present invention relates to a method for encoding at least one image cut into data blocks, implementing, for a current block to be encoded in the image, the current block containing M line vectors and N column vectors of data, such as M> 2 and N> 2, processing the data of the current block by applying a transform which is selected according to a predetermined coding performance criterion, in a set comprising at least two transforms, a transform of the set implementing the following operations: - first data processing applied on the M or N vectors of the current block, at the end of which a data block is obtained, - second data processing applied on the N or M vectors of the processed data block, at the end of which is obtained a transformed data block.

Such a coding method is remarkable in that at least one other transform of the set implements: either only a processing of the M data line vectors of the current block, or only a processing of the N column data vectors. of the current block.

By taking into account, for the purpose of selecting a transform, at least one transform that applies to only the M vector data lines of the current block, or only to the N column vectors of the current block , the memory resources at the encoder are advantageously reduced.

Such an arrangement also makes it possible to reduce, in a non-negligible manner, the calculations related to the placing in competition between the different candidate transforms of the set, according to the predetermined coding performance criterion, of each of the proposed transforms.

The selection of the transform to be applied is thus faster.

In addition, in the case where a transform is selected that applies either only to the M vector data lines of the current block, or only to the N column vectors of data of the current block, rather than a transform that implements two data processing for a current block considered, the processing of the data of the current block using the thus selected transform is significantly accelerated.

According to another particular embodiment, a transform of the set, implementing either only a treatment of the M line vectors of the current block, or only a processing of the N column vectors of the current block, is a trigonometric transform.

With respect to the transforms implementing a first and a second data processing, as conventionally used in the current coding / decoding standards, such an arrangement makes it possible to obtain the best compromise between, on the one hand, a significant improvement the gain in coding and, secondly, a low computational complexity on the data of the block which itself results from the implementation of the selection of the transform according to the competition of particular transforms according to the invention.

According to another embodiment, if a transform is selected using either only a processing of the M data line vectors of the current block, or only a processing of the N column vectors of data of the current block, the choice of the processing is a function of the prediction direction of the data of the current block.

Such a disposition advantageously makes it possible to condition the choice of the single line or column type processing to be applied to the data of the current block, to the prediction direction chosen to predict the data of the current block.

Thus, said selected transform is optimally adapted to the prediction mode chosen for the current block, such as in particular the Intra mode. That is to say, for a given bit rate, such an adaptation has the advantage of not causing degradation of the coded data of the current block according to the chosen prediction mode.

The various embodiments or aforementioned embodiments can be added independently or in combination with each other, to the steps of the coding method as defined above. The invention also relates to a device for encoding at least one image divided into blocks, comprising a processing circuit which, for a current block to be encoded in the image, the current block containing M row vectors and N column vectors of data. , such that M> 2 and N> 2, is arranged to process data of the current block by applying a transform that is selected according to a predetermined coding performance criterion, in a set comprising at least two transforms, a transform of the set implementing the following operations: - first data processing applied on the M or N vectors of the current block, at the end of which is obtained a processed data block, - second data processing applied on the N or M vectors of the processed data block, at the end of which is obtained a transformed data block.

The coding device according to the invention is remarkable in that at least one other transform of the set implements: either only a treatment of the M vector data lines of the current block, or only a processing of the N vectors data columns of the current block.

Such a coding device is particularly suitable for implementing the aforementioned coding method. The invention also relates to a method for decoding a data signal representative of at least one image divided into blocks, implementing, for a current block to be decoded from the image, the current block containing M line vectors and N data column vectors, such as M> 2 and N> 2, as follows: - determining, in the data signal: • a current block of coded data associated with the current block to be decoded, • an indicator of a transform to be applied to the data of the current block of coded data, - processing of the coded data of the current block by applying to the coded data of the transform associated with the determined indicator, such a transform being selected from a set of transforms comprising at least two transforms, a transform of the set implementing the following operations: • first data processing applied on the M or N vectors of the current block, at the end of which is obtained a data block processed, • second data processing applied to the N or M vectors of the processed data block, after which a transformed data block is obtained.

Such a decoding method is remarkable in that at least one other transform of the set implements: either only a treatment of the M coded data line vectors of the current block, or only a processing of the N column vectors of encoded data of the current block.

According to a particular embodiment, a transformation of the set, implementing either only a processing of the M line vectors of the current block of coded data, or only a processing of the N column vectors of the current block of coded data, is a transform. trigonometric.

According to another particular embodiment, if, following the determination of the indicator of the transform to be applied to the coded data of the current block, a transform is selected using only the M coded data line vectors. current block, that is only a processing of N coded data column vectors of the current block, the choice of the processing is a function of the prediction direction of the coded data of the current block.

The various embodiments or aforementioned embodiments can be added independently or in combination with each other, to the steps of the decoding method as defined above. The invention also relates to a device for decoding a data signal representative of at least one image divided into blocks, comprising a processing circuit which, for a current block to be decoded from the image, the current block containing M vectors. rows and N column vectors of data, such as M> 2 and N> 2, are arranged to: - determine, in the data signal: • a current block of coded data associated with the current block to be decoded, • an indicator of a transform to be applied to the data of the current block of coded data, - processing coded data of the current block by applying to the coded data of the transform associated with the determined indicator, the transform being selected from a set of transforms comprising at least two transforms , a transform of the set implementing the following operations: • first data processing applied on the M or N vectors of the current block, at the end of which l is obtained a processed data block, • second data processing applied to the N or M vectors of the processed data block, at the end of which is obtained a transformed data block.

The decoding device according to the invention is remarkable in that at least one other transform of the set implements: either only a processing of the M coded data line vectors of the current block, or only a processing of the N encoded data column vectors of the current block.

Such a decoding device is particularly suitable for implementing the aforementioned decoding method. The invention also relates to a computer program comprising instructions for implementing one of the coding and decoding methods according to the invention, when it is executed on a computer.

This program can use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other form desirable shape. The invention also relates to a computer-readable recording medium on which a computer program is recorded, this program comprising instructions adapted to the implementation of one of the coding or decoding methods according to the invention. as described above. The invention also relates to a recording medium readable by a computer on which a computer program is recorded, this program including instructions adapted to the implementation of the coding or decoding method according to the invention, as described. above.

The recording medium may be any entity or device capable of storing the program. For example, the medium may comprise storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, a USB key, or a magnetic recording means, for example a hard disk. On the other hand, the recording medium may be a transmissible medium such as an electrical or optical signal, which may be conveyed 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.

Alternatively, the recording medium may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the aforementioned coding or decoding method.

BRIEF DESCRIPTION OF THE DRAWINGS Other characteristics and advantages will appear on reading a preferred embodiment described with reference to the figures in which: FIG. 1 represents the steps of the coding method according to the invention, FIG. an embodiment of a coding device according to the invention; FIG. 3 represents an example of a current block to be coded / decoded; FIG. 4 represents an embodiment of a decoding device according to the invention; FIG. 5 represents the steps of the decoding method according to the invention.

Detailed description of the coding part

An embodiment of the invention will now be described, in which the coding method according to the invention is used to code an image or a sequence of images according to a bit stream close to that obtained by a conforming coding. for example to the HEVC standard.

In this embodiment, the coding method according to the invention is for example implemented in a software or hardware way by modifications of an encoder initially conforming to the HEVC standard. The coding method according to the invention is represented in the form of an algorithm comprising steps C1 to C7 as represented in FIG.

According to the embodiment of the invention, the coding method according to the invention is implemented in a coding device CO represented in FIG.

As illustrated in FIG. 2, such an encoder device comprises: an input ENT_C for receiving a current image to be encoded; a processing circuit CT_C for implementing the coding method according to the invention, the processing circuit CT_C containing: A memory MEM_C comprising a buffer memory MT_C, a processor PROC_C controlled by a computer program PG_C, an output SOR_C to deliver a signal or coded stream containing the data obtained at the end of the coding of the current image. At initialization, the code instructions of the computer program PG_C are for example loaded into a RAM memory, MR_C, before being executed by the processing circuit CT_C.

The coding method represented in FIG. 1 applies to any current image ICj that is fixed or part of a sequence of L images ICi, ..., ICj, ..., ICl (1 <j <L) to code.

During a step C1 represented in FIG. 1, in a manner known per se, a current image ICj is partitioned into a plurality of blocks B-1, B 2,. .., Bf (1 <i <F), a current block selected from this plurality containing M row vectors and N column data vectors, such that M> 2 and N> 2. Such a partitioning step is implemented by a partitioning device MP_C shown in FIG. 2, which device is controlled by the processor PROC_C.

It should be noted that for the purposes of the invention, the term "block" means coding unit (coding unit). This last terminology is notably used in the standard HEVC "ISO / IEC / 23008-2 Recommendation ITU-T H.265 High Efficiency Video Coding (HEVC)".

In particular, such a coding unit groups together sets of pixels of rectangular or square shape, also called blocks or macroblocks.

Such a coding unit could, in a future standard, also group together sets of pixels having other geometric shapes.

Said blocks Bi, B2,..., B ,,..., Bf are intended to be coded according to a predetermined order of travel, which is for example of the lexicographic type. This means that the blocks are coded one after the other, from left to right, then from top to bottom. Other types of course are of course possible. Thus, it is possible to cut the image ICj into several subimages called slices and to independently apply a division of this type on each sub-image. It is also possible to code not a succession of lines, as explained above, but a succession of columns. It is also possible to browse the rows or columns in one direction or the other.

Each block can also be divided into sub-blocks which are themselves subdividable.

During a step C2 shown in FIG. 1, the coder CO selects as current block a first block to be coded B, of the image ICj, such as for example the first block Bi.

An example of such a block is shown in FIG. 3. It contains, for example: M = 8 data line vectors, such as a first line vector Mi of four data di, d2, d3, d4, a second line vector M2 of four data d5, d6, d7, d8, a third line vector M3 of four data dg, di0, du, di2, a fourth line vector M4 of four data di3, of, di5, di6, a fifth line vector M5 of four data d-i7, dis, dig, d2o, a sixth line vector M6 of four data d2i, d22, d23, d24, a seventh line vector M7 of four data d2s, d26, d27, d28, an eighth line vector M8 of four data d2g, d30, d3i, d32, - and N = 4 column vectors, such as a first column vector

Neither of eight data di, ds, dg, di3, di7, d2i, d25, d29, a second column vector N2 of eight data d2, d6, di0, d1, d8, d22, d26, d30, a third column vector N3 of eight data d3, d7, di5, dig, d23, d27, d3i, a fourth column vector N4 of eight data d4, d8, di2, d6, d2o, d24, d28, d32.

During a step C3 represented in FIG. 1, the coder CO selects a transform from a set E of transforms previously stored in the buffer memory MT_C of the coder CO of FIG. 2.

According to the invention, the set E comprises at least two transforms T0 and Tutelles that: - one of the at least two transforms, for example To, implements the following operations: • first data processing applied on the M or N vectors of the current block B ,, at the end of which is obtained a processed data block Β °, • second data processing applied to the N or M vectors of the processed data block Β °, at the end of which is obtained a transformed data block Btj, - at least one other of the at least two transforms of the set E, here T-ι, implements: • either only a processing of the M vector data lines of the current block, • or only a processing of N column vectors of data of the current block.

The selection of the To or Ti transform is carried out according to a predetermined coding performance criterion, for example: by minimizing the rate / distortion criterion well known to those skilled in the art, or by minimizing only the flow rate; or by minimizing only the distortion, or by taking into account the complexity generated by the computation complexity of the transform when it is selected, said complexity being weighted with the aforementioned rate / distortion criterion.

Recall that in a video coding context, the complexity is defined for example by counting the number of mathematical operations (including addition, multiplication, binary shift) involved for calculating the data transform of the current block B,.

In the example described here, the term "data" means the pixels of the current block B 1.

It should be noted, however, that data are also understood to mean the pixels of a predicted block obtained by means of a prediction of the current block B, with respect to a predictor block which is selected following a call for competition. different prediction modes inter, intra or other predetermined, for example by minimizing a distortion rate criterion.

Transform T0 is composed of a pair of transforms consisting of a primary transform D0 and a secondary transform C0.

According to a first embodiment: the primary transform D0 is a 4x4 matrix which is applied on the M = 8 line vectors of the current block B ,, so as to obtain a treated current block B ^ Do-B *, where t represents the transpose of the block B ,, - the secondary transform Co is an 8x8 matrix which applies to the N = 4 column vectors of the treated current block Β °, so as to obtain a transformed current block ΒΤΐ = Ο0. (Β ° ί) 1.

According to a variant of this first embodiment: the primary transform Cd is an 8x8 matrix which applies to the N = 4 column vectors of the current block B ,, so as to obtain a treated current block B ° j = Cd. Bi, - the secondary transform C0 is a 4x4 matrix which applies to the M = 8 line vectors of the current processed block Β °, so as to obtain a transformed current block ΒΤΐ = Οο · (Β ° ί) 1.

According to a second embodiment, the transforms Co and D0 are applied through a fast implementation by a fast algorithm that can take the form of butterflies (butterfly diagram in English). This applies in particular to trigonometric transforms, as described in the publication Algebraic Signal Processing Theory: Cooley-Tukey Type Algorithms for DCTs and DSTs, IEEE Transactions on Signal Processing, April 2008.

According to this second embodiment, the transformed current block BT is obtained in less arithmetic operations than by successively applying the primary and secondary transforms D0 and C0 in the form of a matrix product.

In addition, according to the invention, the primary transform D is, for example, of the distortion rate optimized type according to the RDOT (Spleen-Distortion Optimized Transform) criterion, in accordance with the publication OG Sezer et al., "Sparse Orthonormal Transforms for Image Compression ", IEEE ICIP, pp. 149-152, 2008.

According to one embodiment of the invention, the transform Ti of the set E is a trigonometric transform.

According to another embodiment of the invention, in the case where the current block B is predicted according to a given prediction mode, Intra for example, which corresponds to a given prediction direction: - if the To transform is selected as optimal transform T * at the end of step C3, the order in which are treated first M line vectors then N column vectors of the current block B ,, first N column vectors then the M line vectors of the current block B, is a function of the prediction direction of the data of the current block which is associated with the given prediction mode, - if the transform Ti is selected as the optimal transform T * to the from step C3, the choice to process either the M row vectors or the N column vectors of the current block B is a function of the prediction direction of the data of the current block which is associated with the given prediction mode.

During a step C4 shown in FIG. 1, the selected transform T * is applied to the current block B 1. At the end of step C4 is obtained a transformed data block BT ,.

Such an operation is performed by a transform calculation device MTR_C, as shown in FIG. 2, which device is controlled by the processor PROC_C.

During a step C5 shown in FIG. 1, the data of the transformed block BT are quantized according to a conventional quantization operation, such as, for example, a scalar or vector quantization. A block Bq of quantized coefficients is then obtained. Step C5 is implemented by a quantization device MQ_C as represented in FIG. 2, which device is controlled by the processor PROC_C.

In a manner known per se, during a step C6 shown in FIG. 1, the data of the block Bq 1 is encoded. Such coding is, for example, entropic coding of CABAC type ("Context Adaptive Binary Arithmetic Coder" in English) or else an entropy coding of arithmetic type or Huffman type. At the end of step C6, coded data associated with the current block B 1 are obtained. Step C6 is implemented by a coding device MC_C shown in FIG. 2, which device is controlled by the processor PROC_C.

During a step C7 represented in FIG. 1, a signal or data stream φ is constructed which contains the coded data of the current block B, obtained at the end of the aforementioned step C6. Step C7 is implemented by a data signal construction device MCF, as shown in FIG. 2, which device is controlled by the processor PROC_C.

The data signal φ is then delivered via the output SOR_C of the encoder CO of FIG. 2. Such a signal is either stored in the buffer memory MT_C of the encoder CO of FIG. 2, or transmitted by a communication network (not represented). to a remote terminal. This includes the decoder DO shown in FIG.

In a manner known per se, the data signal φ furthermore comprises certain information encoded by the coder CO, such as the type of prediction (Inter or Intra) that has possibly been applied, and if appropriate, the prediction mode selected. the index of the predictor block selected, the reference image index and the motion vector used in the Inter prediction mode, an IT * index associated with the transform T * applied during the aforementioned step C4.

Class B is then decoded conventionally. A decoded BD block is then obtained. It should be noted that the decoded block BD is the same as the decoded block obtained at the end of the image decoding process ICj which will be described later in the description. The decoded block BD is thus made available for use by the coder CO of FIG. 2.

The coding steps C1 to C7 which have just been described above are then implemented for each of the blocks Bi, B2,..., B ,,..., Bf to be encoded of the current image ICj considered. in a predetermined order which is for example the lexicographic order.

Various possible configurations of sets of transforms will now be described.

It is for example considered that the CO encoder of FIG. 2 has a set of two transforms for each proposed prediction mode and for each block size. In this context, and in accordance with the publication "Mode-dependent transform competition in HEVC", Adrià Arrufat et al., Image Processing (ICIP), IEEE ICIP 2015, pp. 1598-1602, the coder CO can thus, during the above-mentioned step C3, put in competition a plurality of transforms by prediction mode.

Thus, for example, for a current block B, of size 4x4 and for a given prediction, for example the HEVC Intra 6 prediction mode, the set E comprises two transforms T0 and Ti which are put into competition during of step C3 of FIG.

According to a first embodiment of this configuration, and as already described above, the transform To is composed of a pair of primary transforms D0 and secondary Co, which are for example of the same type, namely each a type VII DST . The coefficients of each of the transforms C0 and D0 are thus expressed in the form of the equation below, to a multiplicative factor of:

where n is a spatial index (abscissa), such that 0 ^ n <3, and k is a frequency index (ordinate), such that 0 <k <3. The coefficients of each of the transformations C0 and D0 obtained can be approximated and expressed as, for example, whole numbers below, after each multiplied by a multiplicative factor equal to 27: 29 55 74 84 74 74 0 -74 84 -29 -74 55 55 -84 74 -29

A matrix of this type is considered quasi-orthogonal, which advantageously implies that its inverse is its transpose. As such, the product of such a matrix by its transpose gives a matrix having on each of its lines only negligible terms except one, significantly larger than the others (for example a factor 100).

According to this first embodiment, the transform Ti is itself a transform of the type optimized for distortion rate according to the aforementioned criterion RDOT.

The transform Ti is represented, for example, in the form of following integers, represented on 8 bits: 33 55 74 81 61 73 9 -84 99 -31 -63 38 39 -83 82 -34

The transform Τι is composed of: a D-ι transform of type optimized in distortion rate in accordance with the RDOT criterion, which preferably applies to the M = 4 line vectors of the current residual block Br, resulting from the prediction of the block current B, according to the prediction mode Intra No. 6 of the HEVC standard, so as to obtain a processed current block Β1ί = Τ- | .Βηι, where t represents the transpose of the current residue block Bn, - of an identity matrix ID of 4x4 size which preferably applies to the N = 4 column vectors of the treated current block B1 ,, so as to obtain a transformed current block ΒΤί = Ιϋ. (Β1ί) 1.

The composition of the transform Ti also takes account of the fact that the prediction direction of the prediction mode No. 6 is essentially horizontal.

In a manner known per se, the ID identity matrix is in the following form, with a normalization factor of:

As a result, the transformed current block BT can be written as:

Which amounts to:

In a particularly advantageous manner, the number of arithmetic operations is thus only related to the application of a transform of size in accordance with the number of columns of the current block B 1.

As indicated in the comparative table below, it thus appears that the computational complexity generated following the application of the transform Τι is limited to seven arithmetic operations per pixel, such complexity being: - much lower than that generated following the application of a product-type transform of two matrices, denoted Tpm, which requires fourteen arithmetic operations, - a little lower than that generated following the application of the transform T0 composed of the same two trigonometric transforms DST-VII, which requires eight arithmetic operations.

The coding performances which use the set of transforms E compound, according to the invention, of the transforms To and Ti, are compared with those using a set of transforms composed according to the state of the art, noted EEa, which is for example, composed of the To transform and a T10 transform which is itself composed of a pair of D10 primary and secondary C10 transforms, for example jointly optimized according to a RDOT criterion.

Transforms C10 and D10 are respectively in the form of the following matrices: 38 57 72 79 90 56 -20 -67 66 -50 -69 67 47 -85 76 -30 0-17 127 127 6 2 0 -6 89 91 - 4 -3 91 -89 6

The comparison of the coding performances is carried out in relation with the current block B, of size 4x4 and for the prediction mode Intra No. 6 of the HEVC standard.

The comparison is evaluated for example using a measure compactness in the distortion / parsimony plan, according to the Sezer publication cited above. For this purpose, on a TestSet set of residual image signals, the following quantity J, which must be the lowest possible, is measured:

such that: - Brv is a current residual block which is collected on a large set of varied images, - Bqv is the quantized current residual block obtained after application of the considered transform Tu on the residual current block Brv, such that u = {0,1,10}, - ||. || o represents the zero norm, that is the number of non-zero coefficients of the residual residual Bqv block quantized by thresholding, - λ is a Lagrangian operator weighting which adjusts the transmission rate constraint, - R is the transmission rate of the residual current quantized Bqv block, - D is the distortion of the residual current quantized Bqv block, - G is the number of residual blocks considered. , - K is the number of pixels of each residual block.

The table below summarizes the values of R, D and J that are obtained for each of the sets E and EEa proposed in the example above.

It is noted that there is a small difference in performance between the set E and the set Eea to the benefit, for the set E, of a lower computational complexity, since the transform Ti requires seven operations per pixel, whereas Transform T10 requires fourteen operations per pixel.

According to a second embodiment of the aforementioned configuration, the set E comprises the above-mentioned transform T0 and another type of transform T ^ than that mentioned above, the transform Ti being still an optimized transformation in distortion flow according to the criterion RDOT.

In this second embodiment, the transform Ti is a discrete trigonometric transform DTT (abbreviation of "Discrete Trigonometry Transforms").

The transform Ti is composed of: a transform D 1 of type optimized in distortion rate according to the criterion RDOT, which applies for example to the M = 4 line vectors of the current residue block Βη, so as to obtain a treated current block Β1, = Τι.Βηι, where t represents the transpose of the current residual block Br ,, - of a identity matrix ID of size 4x4 which applies to the N = 4 column vectors of the treated current block B1 ,, so as to obtain a transformed current block B ^ ID ^ B1,) 1.

According to this second embodiment, the optimized transform according to the RDOT criterion is a DCT-V transform which is expressed in the following form: 48 68 68 68 68 60 -22 -87 68 -22 -87 60 68 -87 60 - 22

In a particularly advantageous way, the number of arithmetic operations associated with the transform Ti is therefore only related to the application of a transform of size in accordance with the number of columns of the current block Bi.

The coding performances which use the set of transforms E composed, according to the second embodiment of the invention, of the transforms T0 and Τι, are compared with those using a set Eea of transforms composed according to the state of the art, which is for example composed of the transform T0 and of a transform T20 which is itself composed of a pair of primary transforms D20 and secondary C2o, which are for example each trigonometric and jointly optimized according to a criterion RDOT.

The optimized primary D20 transform is a DST-VII transform already shown above in the description.

The optimized secondary transform C2o is, for example, a DCT-IV transform which can be expressed using a formula or be expressed in eight-bit integer values as below:

DCT-IV 89 75 50 18 75 -18 -89 -50 50 -89 18 75 18 -50 75 -89

The comparison of the coding performances is implemented in the same way as in the first embodiment.

The table below summarizes the values of R, D and J that are obtained for each of the sets E and EEa proposed according to this second embodiment.

It is found that the coding performance is much better by using the set E for the benefit of a lower computational complexity since the number of arithmetic operations required by the transform Ti represents approximately half the number of arithmetic operations required by the transformed T2o-

The configuration of the set E of transforms which has been described above is of course variable, in particular according to the prediction mode chosen to predict the current block. Thus, for vertical prediction modes, the transform Ti is rather composed as follows: an identity matrix ID of size 4x4 which preferably applies to the M = 4 line vectors of the current residue block Βη, so as to to obtain a processed current block Β1ί = Ιϋ. (Βη) 1, - a D-type transform optimized in distortion rate in accordance with the RDOT criterion, which preferably applies to the M = 4 column vectors of the treated current block B1 ,, of way to obtain a transformed current block ΒΤί = Τ-ι. (ΒηΥ.

In a particularly advantageous manner, the number of arithmetic operations is thus only related to the application of a transform of size in accordance with the number of rows of the current block B 1.

It is also obvious that the set E of transforms may comprise more than two transforms. Different examples of sets E of transforms which make it possible to obtain good compromises in terms of computational complexity / coding performance, are listed below: E = (T0, Ti, T2), such that T0 = (DST- VII, DST-VII), T1 = (ID, DCT-V) and T2 = (ID, DST-VI); E = (T0, Ti, T2, T3, T5), such that T0 = (DST-VII, DST-VII), T1 = (DCT-IV, DST-VII), T2 = (ID, DCT-II) , T3 = (DCT-VI, DCT-VI), T4 = (ID, DST-IV); E = (T0, T1; T2), such that T0 = (DST-VII, DST-VII), T1 = (DCT-V, ID) and T2 = (DCT-V, DCT-VII); E = (T0, T1, T2, T3, T5), such that T0 = (DST-VII, DST-VII), T1 = (DST-IV, DCT-III), T2 = (DCT-V, ID) , T3 = (DCT-V, DCT-IV), T4 = (DCT-VI, DCT-V).

Detailed description of the decoding part

An embodiment of the invention will now be described, in which the decoding method according to the invention is used to decode a signal or data stream representative of an image or a sequence of images which is suitable for be decoded by a decoder according to any of the current or future video decoding standards.

In this embodiment, the decoding method according to the invention is for example implemented in a software or hardware way by modifications of such a decoder.

The decoding method according to the invention is represented in the form of an algorithm comprising steps D1 to D7 as represented in FIG.

According to this embodiment, the decoding method according to the invention is implemented in a decoding device or decoder DO represented in FIG. 4.

As illustrated in FIG. 4, such a decoder device comprises: an input ENT_D for receiving the data signal or current flow φ to be decoded, a processing circuit CT_D for implementing the decoding method according to the invention, the circuit CT_D processing device containing: a memory MEM_D comprising a buffer memory MT_D, a processor PROC_D controlled by a computer program PG_D, an output SOR_D to deliver a reconstructed current image containing the data obtained after the decoding according to the method of the invention. At initialization, the code instructions of the computer program PG_D are for example loaded into a RAM memory, MR_D, before being executed by the processing circuit CT_D.

The decoding method shown in FIG. 5 applies to a signal or data stream φ representative of a current image ICj to be decoded which is fixed or which belongs to a sequence of images to be decoded. For this purpose, information representative of the current image ICj to be decoded is identified in the data signal φ received at the input ENT_D of the decoder DO and as delivered at the end of the coding method of FIG.

With reference to FIG. 5, during a step D1, it is known, in a manner known per se, to determine in the signal φ coded blocks associated with each of the blocks Bi, B2, B ,, ... , Bf previously encoded according to the above lexicographic order.

Such a determination step D1 is implemented by an identification device MI_D flow analysis, as shown in Figure 4, which is controlled by the processor PROC_D. Other types of course than the one mentioned above are of course possible and depend on the order of course chosen coding.

During a step D2 shown in FIG. 5, the decoder DO of FIG. 4 selects, as current block B, to be decoded, the first block which has been coded at the end of the coding method of FIG.

An example of such a block is shown in FIG. 3. It contains, for example: M = 8 data line vectors, such as a first line vector Mi of four data di, d2, d3, d4, a second line vector M2 of four data ds, d6, d7, d3, a third line vector M3 of four data dg, d-io, dn, di2, a fourth line vector M4 of four data di3, du, di5, di6, a fifth line vector M5 of four data of, dis, di9, d20, a sixth vector line M6 of four data d2i, d22, d23, d24, a seventh vector line M7 of four data d2s, d26, d27, d23, an eighth vector line M8 of four data d2g, d30, d3i, d32, - and N = 4 column vectors, such as a first column vector

Neither of eight data di, ds, dg, di3, di7, d2i, d2s, d2g, a second column vector N2 of eight data d2, d6, d-ιο, du, dis, d22, d26, d3o, a third column vector N3 of eight data d3, d7, di5, d-ig, d23, d27, d3i, a fourth column vector N4 of eight data d4, d3, di2, d-16, d20, d24, d23, d32.

During a step D3 shown in FIG. 5, a determination, for example by decoding, of the data associated with the current block B, to be decoded, which have been coded during the first step, is carried out in a manner known per se. step C6 of FIG. 1. On completion of such a determination, a set of digital information associated with the quantized coefficient block Bq is obtained, which was obtained at the end of the quantization step C5 of FIG. Figure 1.

Also during the step D3, information can be determined which is related to the prediction type of the current block B ,, if the latter has been predicted at the coding, and which has been written in the data signal φ. Such prediction information is in particular the prediction mode selected at the coding and the index of the selected predictor block.

During step D3, the index IT * of the selected transform is also determined, in a manner known per se, at the coding in step C3 of FIG. 1, then applied to step C4 of FIG. .

Such a decoding step D3 is implemented by a decoding device MD_D shown in FIG. 4, which device is controlled by the processor PROC_D.

During a step D4 shown in FIG. 5, the quantized coefficient block Bq 1 is dequantized according to a conventional dequantization operation which is the inverse operation of the quantization implemented during the step for quantization C5 of FIG. 1. A set of current dequantized coefficients BDq is then obtained at the end of step D4. Such a dequantization step is for example of scalar or vector type. Step D4 is implemented by means of an inverse quantization MQ'1_D device, as represented in FIG. 4, which device is controlled by the PROC_D processor.

During a step D5 shown in FIG. 5, the decoder DO proceeds to the selection of a transform T'1 * associated with the index IT * determined in the aforementioned step D3, from a set E'1 of transformed previously stored in the buffer MT_D of the decoder DO of FIG. 4.

According to the invention, the set E'1 comprises at least two transforms T'1o and T'11, respectively inverse of the transforms T0 and Ti used at the encoder, such that: one of the at least two transforms, for example T'10, implements the following operations: • first data processing applied on the M or N vectors of the current dequantized coefficient block BDq ,, at the end of which is obtained a processed data block Β °, • second treatment data set applied on the N or M vectors of the processed data block Β °, at the end of which is obtained a transformed decoded data block BDtj, - at least one other of the at least two transforms of the set E'1, here T'1i, implements: • either only a processing of the M vector data lines of the current dequantized coefficient block BDq ,, • or only a processing of the N column vector data of the current dequantized coefficient block BDqi.

In the example described here, data is understood to mean the pixels of the transformed decoded data block BDtj.

It should be noted, however, that data are also understood to mean the pixels of a modified decoded residue transformed block, in the case where a prediction of the current block B has been implemented at the coding.

Transform T10 is composed of a pair of transforms consisting of a primary transform D'10 and a secondary transform C'10 respectively inverse of the primary transforms D0 and secondary Co mentioned above.

According to a first embodiment: the primary transform D'10 is a 4x4 matrix which applies to the M = 8 line vectors of the current dequantized coefficient block BDq ,, so as to obtain a treated current block BVD'ViBDqi) 1, where t represents the transpose of the block BDq ,, - the secondary transform C'1o is an 8x8 matrix which applies to the N = 4 column vectors of the processed current block Β °, so as to obtain a transformed decoded current block ΒΟΤΐ = Ο'10. (° Β ί) 1.

According to a variant of this first embodiment: the primary transform D'10 is an 8 × 8 matrix which applies to the N = 4 column vectors of the current dequantized coefficient block BDq ,, so as to obtain a current treated block B ° i = Of 1o. (BDqi), - the secondary transform C'10 is a 4x4 matrix that applies to the M = 8 line vectors of the current processed block Β °, so as to obtain a decoded current block transformed BDTi ^ 'ViB0,) 1.

According to a second embodiment, correspondingly to the above coding, the C'10 and D'10 transforms are applied through a fast implementation by a butterfly algorithm.

According to this second embodiment, the transformed decoded current block BDI is obtained in less arithmetic operations than by successively applying the matrices of primary and secondary transforms D 1o and CV

According to one embodiment of the invention, the T'1i transform of the set E'1 is a trigonometric transform.

According to another embodiment of the invention, in the case where a prediction of the current block B has been implemented in the coding according to a given prediction mode, Intra for example, which corresponds to a prediction direction given. if the transform T10 is selected at the end of the step D5, the order in which the M line vectors and then the N column vectors of the current dequantized coefficient block BDq are first processed, or first the N column vectors and then the M line vectors of the current dequantized coefficient block BDq ,, depends on the prediction direction of the data of the current block which is associated with the given prediction mode, - if the transform T'1i is selected at at the end of step D5, the choice to process either the M row vectors or the N column vectors of the current dequantized coefficient block BDq depends on the prediction direction of the data of the current block which is associated with the given prediction mode.

During a step D6 shown in FIG. 5, the selected transform T1 * is applied to the current dequantized coefficient block BO.sub.q. At the end of step D6 is obtained a transformed decoded data block BDTi which constitutes the reconstructed current BD, BD.

Such an operation is performed by a transform calculation device MTR'1_D, as shown in FIG. 4, which device is controlled by the processor PROC_D.

During a step D7 shown in FIG. 5, said current reconstructed block BD is written in a decoded image IDj.

Such a step is implemented by an image reconstruction URI device as shown in FIG. 4, said device being controlled by the PROC_D processor.

The decoding steps D1 to D7 which have just been described above are implemented for all the blocks B-ι, B2,..., B ,,..., BF to be decoded from the current image ICj. considered, in a predetermined order which is for example the lexicographic order.

It goes without saying that the embodiments which have been described above have been given for purely indicative and non-limiting purposes, and that many modifications can easily be made by those skilled in the art without departing from the scope. of the invention.

Claims

A method for encoding at least one image (ICj) cut into data blocks, implementing, for a current block (B,) to be encoded of said image, said current block containing M row vectors and N column vectors of data, such as M> 2 and N> 2, processing the data of the current block by applying (C4) a transform which is selected according to a predetermined coding performance criterion, in a set (E) comprising at least two transformed, a transform of said set implementing the following operations: - first data processing applied to the M or N vectors of the current block, at the end of which is obtained a processed data block, - second data processing applied to the N or M vectors of the processed data block, at the end of which is obtained a transformed data block, the coding method being characterized in that at least one other transform of said set implements: a treatment of the M vector rows of data of the current block, or only a processing of N column vectors of data of the current block.

2. Coding method according to claim 1, wherein a transform of said set, implementing either only a treatment of the M line vectors of the current block, or only a processing of the N column vectors of the current block, is a trigonometric transform.

3. coding method according to claim 1 or claim 2, wherein si is selected a transform implementing either only a treatment of the M vector rows of data of the current block, or only a processing of N column data vectors of the block current, the choice of the treatment depends on the direction of prediction of the data of the current block.

4. Device (CO) encoding at least one image (ICj) cut into blocks, comprising a processing circuit (CT_C) which, for a current block (B,) to be encoded of said image, said current block containing M row vectors and N data column vectors, such as M> 2 and N> 2, are arranged to process data of the current block by applying a transform which is selected according to a predetermined coding performance criterion, in a set ( E) comprising at least two transforms, a transform of said set implementing the following operations: - first data processing applied to the M or N vectors of the current block, at the end of which is obtained a processed data block, - second data processing applied on the N or M vectors of the processed data block, at the end of which is obtained a transformed data block, characterized in that at least one other transform of said set implements: - either only ent a processing of the M vector rows of data of the current block, or only a processing of the N column vectors of data of the current block.

A computer program comprising program code instructions for performing the steps of the encoding method according to any one of claims 1 to 3, when said program is executed on a computer.

A computer-readable recording medium on which a computer program is recorded comprising program code instructions for performing the steps of the encoding method according to any one of claims 1 to 3, when said program is running on a computer.

7. A method of decoding a data signal (φ) representative of at least one image (ICj) cut into blocks, implementing, for a current block (B,) to be decoded from said image, said current block containing M row vectors and N column vector data, such as M> 2 and N> 2, as follows: - determining, in said data signal: • a current block of coded data associated with the current block to be decoded, • an indicator (IT *) of a transform to be applied to the data of the current block of coded data, - processing of the coded data of the current block by application (D5) to said coded data of the transform associated with the determined indicator, said transformed being selected from a set (E'1) of transforms comprising at least two transforms, a transform of said set implementing the following operations: • first data processing applied to the M or N vectors of the current block, at least a result of which is obtained a processed data block, • second data processing applied to the N or M vectors of the processed data block, after which a transformed data block is obtained, the decoding method being characterized in that at least one other transform of said set implements: either only a processing of the M coded data line vectors of the current block, or only a processing of the N column vectors of coded data of the current block.

The decoding method according to claim 7, wherein a transform of said set, implementing either only a processing of the M line vectors of the current block of coded data, or only a processing of the N column vectors of the current block of coded data, is a trigonometric transform.

The decoding method according to claim 7 or claim 8, wherein if, following the determination of the indicator of the transform to be applied to the coded data of the current block, a transform implementing only one processing is selected. M coded data line vectors of the current block, ie only a processing of the N coded data column vectors of the current block, the choice of processing is a function of the prediction direction of the coded data of the current block.

10. Device (DO) for decoding a data signal (φ) representative of at least one image (ICj) cut into blocks, comprising a processing circuit (CT_D) which, for a current block (B,) to decoding said image, said current block containing M row vectors and N column data vectors, such as M> 2 and N> 2, is arranged to: - determine, in said data signal: • a current block of associated coded data to the current block to be decoded, an indicator (IT *) of a transform to be applied to the data of the current block of coded data, - processing coded data of the current block by applying to said coded data of the transform associated with the determined indicator , said transform being selected from a set (E'1) of transforms comprising at least two transforms, a transform of said set implementing the following operations: • first data processing applied to the M or N vectors of the current block, at the end of which a processed data block is obtained, • second data processing applied to the N or M vectors of the processed data block, at the end of which is obtained a transformed data block, characterized in at least one other transform of said set implements: either only a processing of the M coded data line vectors of the current block, or only a processing of the N coded data column vectors of the current block.

A computer program comprising program code instructions for performing the steps of the decoding method according to any of claims 7 to 9, when said program is executed on a computer.

A computer-readable recording medium on which a computer program is recorded comprising program code instructions for performing the steps of the decoding method according to any one of claims 7 to 9, when said program is running on a computer.