FR3164086A1 - Method and device for encoding and decoding images - Google Patents

Abstract

The invention relates to a method for decoding a region, called the decoding region, of a signal comprising at least two regions, said decoding region comprising a plurality of samples to be decoded, said decoding method comprising the following steps: the decoding (F22) of a group of latent value maps representative of said decoding region of said signal, the obtaining (F20) of a mask of said decoding region in said signal, the decoding (F26) of a set of parameters representative of a neural network, called the synthetic neural network comprising at least one synthetic neural layer, the processing (F27) of said group of latent maps decoded by said synthetic neural network to produce at least said decoding region as output of said synthetic neural network, said processing comprising the application of said at least one synthetic neural layer to transform at least one group of input latent maps, called input maps, into at least one group of latent maps of output, referred to as output maps, said application comprising obtaining a characteristic region of said area to be decoded in said input maps from said mask, said characteristic region comprising the points associated with said area to be decoded and, for at least one point of said area to be decoded, referred to as a decode point, the processing of a point associated with said decode point in at least one of the latent input maps by said neural layer as a function of its proximity to a boundary of said characteristic region. Figure for the abbreviation: Fig. 7

Description

Method and device for encoding and decoding images

The invention relates to the general field of coding one-dimensional or multidimensional signals. It relates more particularly to the compression of digital images or videos.

Digital videos are generally encoded using source code to compress them and reduce the resources required for transmission and/or storage. Numerous encoding standards exist, such as those of the ITU/MPEG organizations (H.264/AVC, H.265/HEVC, H.266/VVC, etc.) and their extensions (MVC, SVC, 3D-HEVC, etc.). In these approaches, image encoding is typically performed by predicting pixels using previously encoded and decoded pixels present in the image being encoded—in which case it is called "intra-prediction"—or previously encoded images—in which case it is called "inter-prediction."

In addition to these traditional approaches, approaches based on artificial intelligence, and in particular neural networks, are tending to develop.

Some neural network approaches use an input signal, such as an image, to train a synthetic neural network on characteristic vectors associated with a specific position in a sample of the input signal to be encoded. These characteristic vectors are constructed from feature maps, which can be at the same resolution as the input signal or at a lower resolution. During training, or construction, the neural network parameters and the feature map values are updated based on a performance metric, such as data rate-distortion. Once training is complete, meaning the performance measurement is satisfactory, the actual encoding of the synthetic neural network parameters and feature map values can be performed and stored or transmitted to the decoder. The encoded signal is then decoded by applying the synthetic neural network to the feature maps.

One drawback of classical and previously described neural approaches is that they do not allow the creation of independently decodable areas in the coded signal, which limits not only the ability to interact with the semantic content of the coded signal but also the possibility of limiting the memory required at the decoder level or parallelizing the decoding of this coded signal.

Therefore, there is a need for a solution that improves upon the classical and neural approaches described above.

• l’obtention d’un premier groupe d’au moins une carte de valeurs latentes représentatif dudit signal,
• l’obtention d’un masque dans ledit signal de ladite zone à coder,
• pour au moins un échantillon de ladite zone à coder, dit échantillon courant, associé à une position dans ledit signal à coder :
  • la construction d’un vecteur caractéristique à partir desdites cartes de valeurs latentes dudit premier groupe, en fonction de ladite position dudit échantillon courant,
  • le traitement dudit vecteur caractéristique par un réseau de neurones artificiels, dit réseau de neurones de synthèse, ledit réseau de neurones de synthèse étant défini par un ensemble de paramètres et comprenant au moins une couche neuronale de synthèse pour obtenir, en sortie dudit réseau de neurone de synthèse, un vecteur représentatif d’une valeur décodée de l’échantillon courant,
  • la mise à jour d'au moins une valeur d’une desdites cartes de valeurs latentes dudit premier groupe et/ou d’au moins un paramètre dudit réseau de neurones de synthèse, en fonction d’une mesure de performance de codage,
• l’obtention d’un deuxième groupe de cartes de valeurs latentes représentatives d’au moins ladite zone à coder à partir dudit masque,
• le codage du deuxième groupe de cartes de valeurs latentes,
• le codage dudit masque, et
• le codage dudit ensemble de paramètres dudit réseau de neurones de synthèse.

ledit traitement comportant l’application de ladite au moins une couche neuronale de synthèse pour transformer au moins un groupe de cartes latentes d’entrée, dite cartes d’entrée, en au moins un groupe de cartes latentes de sortie, dite cartes de sortie, ladite application comprenant :

• l’obtention d’une région caractéristique de ladite zone à décoder dans lesdites cartes d’entrée à partir dudit masque, ladite région caractéristique comprenant les points associés audit masque,
• pour au moins un point de ladite zone à décoder, dit point à décoder :
  • l’obtention d’un point associé audit point à décoder dans au moins une des cartes latentes d’entrée,
  • l’évaluation d'un critère de proximité dudit point associé par rapport à une frontière de ladite région caractéristique,
  • la construction d’un vecteur d’entrée de ladite couche neuronale de synthèse à partir d’un voisinage dudit point associé et de ladite évaluation dudit critère de proximité, et
  • le traitement dudit vecteur d’entrée par ladite couche neuronale de synthèse pour obtenir un point d’au moins une des cartes de sortie,

The invention relates to a method for encoding a region of a signal, called the coding region, said coding region comprising a plurality of samples to be encoded, said coding method comprising the following steps:

• obtaining a first group of at least one latent value map representative of said signal,
• obtaining a mask in said signal from said zone to be coded,
• for at least one sample of said area to be coded, called the current sample, associated with a position in said signal to be coded:
  • the construction of a characteristic vector from said latent value maps of said first group, as a function of said position of said current sample,
  • the processing of said characteristic vector by an artificial neural network, called a synthetic neural network, said synthetic neural network being defined by a set of parameters and comprising at least one synthetic neural layer to obtain, at the output of said synthetic neural network, a vector representing a decoded value of the current sample,
  • the updating of at least one value of one of said latent value maps of said first group and/or of at least one parameter of said synthetic neural network, as a function of a coding performance measure,
• obtaining a second set of latent value maps representative of at least the said area to be encoded from said mask,
• the encoding of the second group of latent value maps,
• the encoding of said mask, and
• the coding of said set of parameters of said synthetic neural network.

said processing comprising the application of said at least one neural synthesis layer to transform at least one group of input latent maps, referred to as input maps, into at least one group of output latent maps, referred to as output maps, said application comprising:

• obtaining a characteristic region of said area to be decoded in said entry maps from said mask, said characteristic region comprising the points associated with said mask,
• for at least one point in said area to be decoded, called point to be decoded:
  • obtaining a point associated with said point to be decoded in at least one of the latent input maps,
  • the evaluation of a proximity criterion of said associated point in relation to a boundary of said characteristic region,
  • the construction of an input vector for said neural synthesis layer from a neighborhood of said associated point and said evaluation of said proximity criterion, and
  • the processing of said input vector by said neural synthesis layer to obtain a point on at least one of the output maps,

• le décodage d’un groupe de cartes de valeurs latentes représentatives de ladite zone à décoder dudit signal,
• l’obtention d’un masque de ladite zone à décoder dans ledit signal,
• le décodage d’un ensemble de paramètres représentatifs d’un réseau de neurones, dit réseau de neurones de synthèse comprenant au moins une couche neuronale de synthèse,
• le traitement dudit groupe de cartes latentes décodées par ledit réseau de neurone de synthèse pour produire en sortie dudit réseau de neurone de synthèse au moins ladite zone à décoder, ledit traitement comportant l’application de ladite au moins une couche neuronale de synthèse pour transformer au moins un groupe de cartes latentes d’entrée, dite cartes d’entrée, en au moins un groupe de cartes latentes de sortie, dite cartes de sortie, ladite application comprenant l’obtention d’une région caractéristique de ladite zone à décoder dans lesdites cartes d’entrée à partir dudit masque, ladite région caractéristique comprenant les points associés à ladite zone à décoder et pour au moins un point de ladite zone à décoder, dit point à décoder, le traitement d’un point associé audit point à décoder dans au moins une des cartes latentes d’entrée par ladite couche neuronale en fonction de sa proximité d'une frontière de ladite région caractéristique.

The invention also relates to a method for decoding a zone, called the zone to be decoded, of a signal comprising at least two zones, said zone to be decoded comprising a plurality of samples to be decoded, said decoding method comprising the following steps:

• the decoding of a group of latent value maps representative of said area to be decoded of said signal,
• obtaining a mask of said area to be decoded in said signal,
• the decoding of a set of parameters representative of a neural network, called a synthetic neural network comprising at least one synthetic neural layer,
• the processing of said group of latent maps decoded by said synthetic neural network to produce at the output of said synthetic neural network at least said area to be decoded, said processing comprising the application of said at least one synthetic neural layer to transform at least one group of input latent maps, said input maps, into at least one group of output latent maps, said output maps, said application comprising obtaining a characteristic region of said area to be decoded in said input maps from said mask, said characteristic region comprising the points associated with said area to be decoded and for at least one point of said area to be decoded, said decode point, the processing of a point associated with said decode point in at least one of the input latent maps by said neural layer as a function of its proximity to a boundary of said characteristic region.

For the purposes of this invention, encoding, or "coding " , refers to the operation of representing a set of samples in a compact form, for example, using a digital binary stream. Decoding refers to the operation of processing a digital binary stream to produce decoded samples.

By "sample" "Sampling" refers to a value taken from the signal. Sampling the signal produces a sequence of discrete values called samples. In the case of an image signal, the sample is called a pixel, which can be, for example, a color pixel traditionally represented by a triplet of values, such as (R, G, B) or (Y, U, V). The position of the sample is located by its x-coordinate (abscissa) and y-coordinate (ordinate) in the image.

A "signal comprising a plurality of samples" refers to a signal with one (audio, sound), two (image), or more than two (stereoscopic, multiscopic, depth-mapped, video, etc.) dimensions. Depending on its dimensionality, the sample has one, two, or more coordinates within the signal. In the case of an image signal, the sample's position is identified by its x-coordinate (abscissa) and y-coordinate (ordinate).

By "feature maps" or equivalently by "latent value maps", we mean an abstract representation of the signal comprising a plurality of variable data, discrete or not, which are also called values, for example real or integer numbers.

By "characteristic data vector constructed from feature maps as a function of a The term "position" refers to a vector consisting of one or more elements, or data points, preferably discrete. These data points are constructed from feature maps at a position determined by that of the sample being processed within the signal. This feature vector is the one applied to the input of the synthesis neural network. For example, in the case of a one-dimensional audio signal, such a vector can be constructed from a plurality of values taken from each of the feature maps at the same coordinate as the sample to be encoded or in a neighborhood thereof. In the case of an image, such a vector can be constructed from a plurality of values taken from each of the feature maps at the same x and y coordinates as the sample to be encoded (or decoded) or in a neighborhood thereof. Once these values are taken from the feature maps, they can be processed to form the feature vector, before entering the synthesis neural network, for example by quantization, filtering, interpolation, etc.

By "synthetic neural network", we mean a neural network such as a convolutional neural network, a multilayer perceptron, an LSTM (for "Long Short Term Memory"), etc. The neural network is defined for example by a plurality of layers of artificial neurons and by a set of activation, weighting and addition functions (for example, a layer can calculate y = f (Ax+b), where y and b are vectors of dimension N, x a vector of dimension M, A is a matrix of dimension MxN, and f is the activation function).

By "neural network parameter" we mean one of the values that characterizes the neural network, for example a weight associated with one of the neurons (filter coefficient, weighting, bias, value affecting the functioning of non-linearity, etc.)

By "treatment" By "synthetic neural network," we mean the application of a function expressed by a synthetic neural network to the input characteristic vector to produce an output vector representative of the sample to be encoded (or decoded). This output vector may contain one or more data points representative of the sample.

A "performance measure" is defined as a measurement between at least one value of a sample to be encoded and a decoded value of that sample. The measure might assess, for example, distortion or perceptual error. It can be performed on one or more samples (e.g., a running sample, the current image, etc.). The measure may also include a measurement of throughput, particularly associated with the encoding of the synthetic neural network and/or the encoding of the feature maps in the first group. The measure may be a combined measurement of throughput and distortion through their weighting. As is well known in the state of the art, the value of this measure is generally minimized until a target value is reached.

Generally, the steps of an encoding or decoding process should not be interpreted as being linked to a notion of temporal succession. In other words, the steps can be carried out in a different order than that indicated in the independent encoding or decoding claim, or even in parallel.

The coding method according to the invention encodes a region of a signal from a representation of that signal in the form of feature maps. These feature maps are segmented into characteristic regions whose values are subsequently encoded entropically, independently of one another. Furthermore, a synthetic neural network is trained on all of these regions, taking into account the segmentation into characteristic regions. Thus, it is possible to obtain a coded representation of a region of the original signal that can subsequently be decoded by the synthetic neural network independently of any other part of that signal.

The decoding process (and symmetrically the encoding process) may further include one or more of the following optional features, taken individually or in any technically possible combination.

• l’obtention d’un point associé audit point à décoder dans au moins une des cartes latentes d’entrée,
• l’évaluation d'un critère de proximité dudit point associé par rapport à une frontière de ladite région caractéristique,
• la construction d’un vecteur d’entrée de ladite couche neuronale de synthèse à partir d’un voisinage dudit point associé et de ladite évaluation dudit critère de proximité, et
• le traitement dudit vecteur d’entrée par ladite couche neuronale de synthèse pour obtenir un point d’au moins une des cartes de sortie.

According to a first characteristic, the processing step of said associated point (Pa _n ) comprises:

• obtaining a point associated with said point to be decoded in at least one of the latent input maps,
• the evaluation of a proximity criterion of said associated point in relation to a boundary of said characteristic region,
• the construction of an input vector for said neural synthesis layer from a neighborhood of said associated point and said evaluation of said proximity criterion, and
• the processing of said input vector by said neural synthesis layer to obtain a point of at least one of the output maps.

According to another characteristic, the neighborhood of the associated point is independent of the evaluation of the proximity criterion.

According to another characteristic, the neighborhood of the associated point is selected based on the evaluation of the proximity criterion.

According to another characteristic, the neighborhood of the associated point is selected based on the associated point.

According to another feature, during the construction of the input vector, a component of the input vector is associated with a point in the neighborhood, the value of the component being equal to the value of the associated point if the associated point is a point in the characteristic region and to a replacement value otherwise.

According to another characteristic, the replacement value is dependent on points in the characteristic region or on points associated with the neighborhood belonging to the characteristic region.

According to another characteristic, the replacement value does not depend on the points of the characteristic region.

According to another characteristic, the proximity criterion is a distance, for example a Euclidean distance.

According to another characteristic, the synthetic neuronal layer is a convolutional neuronal layer.

• l’obtention d’un premier groupe d’au moins une carte de valeurs latentes représentatif dudit signal,
• l’obtention d’un masque dans ledit signal de ladite zone à coder,
• pour au moins un échantillon de ladite zone à coder, dit échantillon courant, associé à une position dans ledit signal à coder :
  • la construction d’un vecteur caractéristique à partir desdites cartes de valeurs latentes dudit premier groupe, en fonction de ladite position dudit échantillon courant,
  • le traitement dudit vecteur caractéristique par un réseau de neurones artificiels, dit réseau de neurones de synthèse, ledit réseau de neurones de synthèse étant défini par un ensemble de paramètres et comprenant au moins une couche neuronale de synthèse pour obtenir, en sortie dudit réseau de neurone de synthèse, un vecteur représentatif d’une valeur décodée de l’échantillon courant,
  • la mise à jour d'au moins une valeur d’une desdites cartes de valeurs latentes dudit premier groupe et/ou d’au moins un paramètre dudit réseau de neurones de synthèse, en fonction d’une mesure de performance de codage,
• l’obtention d’un deuxième groupe de cartes de valeurs latentes représentatives d’au moins ladite zone à coder à partir dudit masque,
• le codage du deuxième groupe de cartes de valeurs latentes,
• le codage dudit masque, et
• le codage dudit ensemble de paramètres dudit réseau de neurones de synthèse.

ledit traitement comportant l’application de ladite au moins une couche neuronale de synthèse pour transformer au moins un groupe de cartes latentes d’entrée, dite cartes d’entrée, en au moins un groupe de cartes latentes de sortie, dite cartes de sortie, ladite application comprenant :

• l’obtention d’une région caractéristique de ladite zone à décoder dans lesdites cartes d’entrée à partir dudit masque, ladite région caractéristique comprenant les points associés audit masque,
• pour au moins un point de ladite zone à décoder, dit point à décoder :
  • l’obtention d’un point associé audit point à décoder dans au moins une des cartes latentes d’entrée,
  • l’évaluation d'un critère de proximité dudit point associé par rapport à une frontière de ladite région caractéristique,
  • la construction d’un vecteur d’entrée de ladite couche neuronale de synthèse à partir d’un voisinage dudit point associé et de ladite évaluation dudit critère de proximité, et
  • le traitement dudit vecteur d’entrée par ladite couche neuronale de synthèse pour obtenir un point d’au moins une des cartes de sortie,

Correspondingly, the invention also relates to a coding device for a signal comprising a plurality of samples to be coded, characterized in that said coding device is configured to implement:

• obtaining a first group of at least one latent value map representative of said signal,
• obtaining a mask in said signal from said zone to be coded,
• for at least one sample of said area to be coded, called the current sample, associated with a position in said signal to be coded:
  • the construction of a characteristic vector from said latent value maps of said first group, as a function of said position of said current sample,
  • the processing of said characteristic vector by an artificial neural network, called a synthetic neural network, said synthetic neural network being defined by a set of parameters and comprising at least one synthetic neural layer to obtain, at the output of said synthetic neural network, a vector representing a decoded value of the current sample,
  • the updating of at least one value of one of said latent value maps of said first group and/or of at least one parameter of said synthetic neural network, as a function of a coding performance measure,
• obtaining a second set of latent value maps representative of at least the said area to be encoded from said mask,
• the encoding of the second group of latent value maps,
• the encoding of said mask, and
• the coding of said set of parameters of said synthetic neural network.

said processing comprising the application of said at least one neural synthesis layer to transform at least one group of input latent maps, referred to as input maps, into at least one group of output latent maps, referred to as output maps, said application comprising:

• obtaining a characteristic region of said area to be decoded in said entry maps from said mask, said characteristic region comprising the points associated with said mask,
• for at least one point in said area to be decoded, called point to be decoded:
  • obtaining a point associated with said point to be decoded in at least one of the latent input maps,
  • the evaluation of a proximity criterion of said associated point in relation to a boundary of said characteristic region,
  • the construction of an input vector for said neural synthesis layer from a neighborhood of said associated point and said evaluation of said proximity criterion, and
  • the processing of said input vector by said neural synthesis layer to obtain a point on at least one of the output maps,

• le décodage d’un groupe de cartes de valeurs latentes représentatives de ladite zone à décoder dudit signal,
• l’obtention d’un masque de ladite zone à décoder dans ledit signal,
• le décodage d’un ensemble de paramètres représentatifs d’un réseau de neurones, dit réseau de neurones de synthèse comprenant au moins une couche neuronale de synthèse,
• le traitement dudit groupe de cartes latentes décodées par ledit réseau de neurone de synthèse pour produire en sortie dudit réseau de neurone de synthèse au moins ladite zone à décoder, ledit traitement comportant l’application de ladite au moins une couche neuronale de synthèse pour transformer au moins un groupe de cartes latentes d’entrée, dite cartes d’entrée, en au moins un groupe de cartes latentes de sortie, dite cartes de sortie, ladite application comprenant l’obtention d’une région caractéristique de ladite zone à décoder dans lesdites cartes d’entrée à partir dudit masque, ladite région caractéristique comprenant les points associés à ladite zone à décoder et pour au moins un point de ladite zone à décoder, dit point à décoder, le traitement d’un point associé audit point à décoder dans au moins une des cartes latentes d’entrée par ladite couche neuronale en fonction de sa proximité d'une frontière de ladite région caractéristique.

The invention also relates to a device for decoding a region of a signal comprising at least two regions, referred to as the region to be decoded, said region to be decoded comprising a plurality of samples to be decoded, characterized in that the decoding device is configured to implement:

• the decoding of a group of latent value maps representative of said area to be decoded of said signal,
• obtaining a mask of said area to be decoded in said signal,
• the decoding of a set of parameters representative of a neural network, called a synthetic neural network comprising at least one synthetic neural layer,
• the processing of said group of latent maps decoded by said synthetic neural network to produce at the output of said synthetic neural network at least said area to be decoded, said processing comprising the application of said at least one synthetic neural layer to transform at least one group of input latent maps, said input maps, into at least one group of output latent maps, said output maps, said application comprising obtaining a characteristic region of said area to be decoded in said input maps from said mask, said characteristic region comprising the points associated with said area to be decoded and for at least one point of said area to be decoded, said decode point, the processing of a point associated with said decode point in at least one of the input latent maps by said neural layer as a function of its proximity to a boundary of said characteristic region.

The characteristics and advantages of the coding or decoding process apply in the same way to the coding or decoding device according to the invention and vice versa.

The invention also relates to a computer program on a recording medium, this program being capable of being implemented in a computer or an encoding or decoding device according to the invention. This program includes instructions adapted to the implementation of the corresponding method. This program may use any programming language and be in the form of source code, object code, or code intermediate between source and object code, such as in a partially compiled form, or in any other desirable form.

The invention also relates to a computer-readable information or recording medium containing the aforementioned computer program instructions. The information or recording medium can be any entity or device capable of storing programs. For example, the medium may include a storage means, such as a ROM (e.g., a CD-ROM or a microelectronic circuit ROM), or a magnetic recording means, such as a floppy disk or hard drive, a DNA sequence, or flash memory. Furthermore, the information or recording medium may be a transmissible medium, such as an electrical or optical signal, which can be transmitted via an electrical or optical cable, by radio link, wireless optical link, or by other means.

The program according to the invention can in particular be downloaded onto an Internet-type network.

Alternatively, each information or recording medium can be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of a process according to the invention.

Brief description of the figures

• laFIG. 1représente schématiquement un dispositif de codage selon un mode de réalisation de l’invention,
• laFIG. 2représente schématiquement un dispositif de décodage selon un mode de réalisation de l’invention,
• laFIG. 3illustre un exemple de réseau de neurones artificiels de synthèse utilisé dans le cadre de l’invention,
• laFIG. 4illustre schématiquement un exemple de traitement opéré par une couche neuronale de synthèse du réseau de neurones de laFIG. 3,
• laFIG. 5illustre schématiquement un deuxième exemple de traitement opéré par une couche neuronale de synthèse du réseau de neurones de laFIG. 3.
• laFIG. 6illustre schématiquement un troisième exemple de traitement opéré par une couche neuronale de synthèse du réseau de neurones de laFIG. 3.
• laFIG. 7est un logigramme représentant un exemple de procédé de codage qui peut être mis en œuvre par le dispositif de codage de laFIG. 1,
• laFIG. 8représente de manière illustrée un procédé de codage utilisé dans un mode de réalisation de l’invention,
• laFIG. 9est un logigramme représentant un exemple de procédé de décodage qui peut être mis en œuvre par le dispositif de décodage de laFIG. 2,
• laFIG. 10représente de manière illustrée un procédé de décodage utilisé dans un mode de réalisation de l’invention,
• laFIG. 11est un logigramme représentant une méthode de codage de cartes de caractéristiques qui peut être mis en œuvre par le dispositif de codage de laFIG. 1et par le procédé de codage de laFIG. 7,
• laFIG. 12est un logigramme représentant une méthode de décodage de cartes de caractéristiques qui peut être mis en œuvre par le dispositif de décodage de laFIG. 2et par le procédé de décodage de laFIG. 9.

The invention will be better understood with the aid of the following description, given solely by way of example and made with reference to the accompanying drawings in which:

• there FIG. 1 schematically represents a coding device according to one embodiment of the invention,
• there FIG. 2 schematically represents a decoding device according to one embodiment of the invention,
• there FIG. 3 illustrates an example of a synthetic artificial neural network used in the context of the invention,
• there FIG. 4 schematically illustrates an example of processing performed by a neural synthesis layer of the neural network of the FIG. 3 ,
• there FIG. 5 schematically illustrates a second example of processing performed by a neural synthesis layer of the neural network of the FIG. 3 .
• there FIG. 6 schematically illustrates a third example of processing performed by a neural synthesis layer of the neural network of the FIG. 3 .
• there FIG. 7 is a flowchart representing an example of a coding process that can be implemented by the coding device of the FIG. 1 ,
• there FIG. 8 represents, in an illustrated manner, a coding method used in one embodiment of the invention,
• there FIG. 9 is a flowchart representing an example of a decoding process that can be implemented by the decoding device of the FIG. 2 ,
• there FIG. 10 represents, in an illustrated manner, a decoding process used in one embodiment of the invention,
• there FIG. 11 is a flowchart representing a method for encoding feature cards that can be implemented by the encoding device of the FIG. 1 and by the coding process of the FIG. 7 ,
• there FIG. 12 is a flowchart representing a method for decoding feature cards that can be implemented by the decoding device of the FIG. 2 and by the decoding process of the FIG. 9 .

Detailed description of the invention

There FIG. 1 This schematically represents, according to a first embodiment, an ENC coding device for at least one region of a signal (I(P _n )). In the example described here, the signal I(P _n ) is segmented into a set of J regions, and all regions are coded independently. Alternatively, only one region or just a few regions can be coded independently.

This ENC coding device includes a SEG segmentation module, an SEGC encoding module of the segmentation provided by the SEG segmentation module, a GEN module for generating feature maps, an SE module for transforming feature maps, an XTR module for extracting data from feature maps, a TT module for processing and quantification, an NNSYN module corresponding to a synthetic artificial neural network, an NNC module for coding a neural network capable of encoding the synthetic neural network, an FMC module for coding feature maps, an EVAL module for evaluating coding performance, and a MAJ module for updating.

The ENC coding device can be implemented by means of an electronic device comprising a processor and a memory, not shown; each of the modules mentioned above can then be realized by the cooperation of the processor and computer program instructions stored in the aforementioned memory and designed to perform the functionalities of the module concerned, in particular as described below, when these instructions are executed by the processor.

The ENC coding device of the FIG. 1 The system receives as input a signal consisting of a succession of samples to be encoded, denoted P_{_n} , for example, a temporal sequence of sound samples, or a set of image data denoted I(P _{_n} ). In this second case, the image signal I(P _{_n}) can represent a two-dimensional image, or a plurality of two-dimensional images (video, color components, stereoscopic components, multiscopic components, etc.). P _{_n} designates a sample n of the input signal comprising N samples. In one embodiment, the signal is a color image signal represented by means of at least one two-dimensional representation, such as a pixel matrix, each pixel having a red component R, a green component G, a blue component B, or, alternatively, a luminance component Y and at least one chrominance component U, V. The location of each pixel is defined by its abscissa x and ordinate y coordinates in the image. In one embodiment, the image is in greyscale represented by means of a two-dimensional representation, such as a pixel matrix, each pixel having a greyscale component, or luminance. In this case, the vector representing the pixel is reduced to a single component.

The SEG module performs a segmentation Sg of the sequence of samples to be encoded P _{_n} into J (greater than or equal to 2) zones Z_o ^j . This segmentation operation allows the samples to be grouped into J different homogeneous sets based on one or more predefined criteria. For example, in the case of a temporal sequence of sound samples, the segmentation operation allows the results to be obtained from sequences of units corresponding to silences, noises, phonemes, words, etc. Similarly, in the case of an image signal I(P _{_n} ), the segmentation operation allows the pixels P _{_n} of the image signal to be grouped into J homogeneous zones according to criteria, notably intensity or spatial ones. For example, the segmentation operation can allow the identification in the image signal I(P _{_n} ) of two different zones (J=2), one corresponding to the background of the image and the other to the foreground of the image. The segmentation Sg is for example represented as a set of J masks MZo ^j each allowing to identify a zone Zo ^j , each of the masks being associated with a value different from that associated with the other masks.

The SEGC module performs lossless encoding of the Sg segmentation. This encoding can be achieved by encoding the segmentation map corresponding to the J MZo ^j masks (for example, using the JPEG-LS algorithm defined by the international standard ISO/IEC 14495-1), or alternatively by independently encoding each MZo j mask, or in yet another variant, by independently encoding the contours of the J MZo ^j masks. The encoded segmentation is denoted Sgc.

The NNC module performs a coding simulation, followed by decoding, for the evaluation module.

The GEN feature map generation module is configured to generate a plurality of M feature maps, denoted FM _i .

In one embodiment, the SE module performs a transformation of the first group of FM _i characteristic maps to generate a second group of FMS _i characteristic maps at the same resolution as the input signal.

The optional SE module can perform quantization of the data extracted from this set of M FM _i maps. Recall that quantizing a value refers to mapping that value to a member of a discrete set of possible code symbols. For example, the set of possible code symbols might consist of integers, and the quantization system would simply round a real value to an integer. Alternatively, quantization could involve multiplying by a given value and then rounding. The SE module then performs a transformation of the values in at least one of the feature maps, such as oversampling, interpolation, filtering, etc. After the transformation, a transformed feature map from the second group has the same resolution as the images in the input sequence. Advantageously, according to this method, the feature maps that are coded can be of lower resolution than the images to be coded, while the maps of the second group, which are used to construct the feature vectors, are at the same resolution as the image sequence, which facilitates the extraction of values.

In one embodiment, the SE module is absent; in this case, the values that will be used to construct the characteristic vector are extracted from the first group of characteristic maps.

The XTR module performs value extraction from the feature maps FMS _i (or FM _i depending on one of the embodiments described previously), for a current sample P _n to be encoded, based on its coordinates in the input signal and possibly the segmentation Sg performed by the SEG module. For example, if we want to encode the sample P _n at coordinates (x _n , y _n ) in a region Zo ^j of an input image, the XTR module performs value extraction from the maps at positions imposed by the coordinates (x _n , y _n ) and by the mask MZo ^j of the region Zo ^j .

In one embodiment, the extracted values constitute the vector Z _{_n} . Z _{_n} is an L-tuple, meaning it contains L elements, or data z_{_i} . For example, in one embodiment, L=M, meaning that only one value is extracted for each feature map FM _{_i} . In another embodiment, L>M, meaning that several values are extracted for at least one feature map FM _{_i}. The dimension L of the vector depends on the topology of the NNSYN synthesis neural network and, more specifically, on the topology of the input layer of this NNSYN synthesis neural network.

The vector Z _n of index n refers to the characteristic vector of the pixel P' _n .

In one embodiment, the optional TT module processes the extracted values to generate the vector Z _{_n} . The TT module can quantify the data extracted from the set of feature maps. This processing may include other operations, such as filtering, scaling, etc. In particular, if the SE module is not used and the feature maps in the first group have lower resolutions than the images in the sequence, the TT module can take into account the coordinates of the values in the lower-resolution maps.

Note that at least one of the SE or TT modules must perform a quantification of the feature maps.

The NNSYN module is a synthetic neural network defined by K parameters W _k , capable of processing the input vector Z _n , or L-Tuple, to generate as output a second vector representative of the sample P _n to be coded.

An example of a synthetic neural network is presented later with reference to the FIG. 3 .

It should be noted that the behavior of the synthesis neural network may depend on the Sg segmentation as will be presented in more detail later with reference to figures 4 to 6.

The NNC module performs the encoding of the synthetic neural network, specifically its parameters W_{_k} . During the training, or construction, process of the encoding—that is, until the performance evaluation stage is satisfactory—the NNC module performs a coding simulation, followed by decoding, for the evaluation module. Subsequently, it performs the actual encoding of the synthetic neural network parameters W _{_k} . The encoded parameters are denoted W_{c k} . As is known, the coding simulation can be identical to the actual encoding, or it can approximate it.

The FMC module performs the encoding of the FM _i maps, that is, the values of the characteristic maps in the first group (excluding the maps in the second group, which may optionally result from oversampling by the SE module). During the encoding training, or construction, process—that is, until the performance evaluation step is satisfactory—the FMC module performs a coding simulation, followed by decoding, for the evaluation module. Subsequently, it performs the actual encoding of the FM _i map values.

The FMC module determines, taking into account the Sg segmentation performed by the SEG module, an SgL segmentation of each of the FM _i feature maps into J zones Thus, the FMC module uses the Sg segmentation defined in the domain of the signal to be encoded to obtain a latent domain segmentation of the FM _i feature maps. Obtaining this latent domain segmentation depends on the transformation transform used to go from the domain of the signal to be encoded to the latent domain. Therefore, the SgL segmentation in the latents results from a calculation that simultaneously considers the location of the Sg segmentation zones in the signal to be encoded and the transformation transform. For example, if the latents and the signal to be encoded have the same resolution, then the latent domain segmentation is identical to the segmentation in the domain of the signal to be encoded. As another example, if a latent has a lower resolution than the signal to be encoded, then the boundary between two zones of the segmentation in the domain of that latent lies between the points in the latent domain whose co-located samples in the signal to be encoded belong to two different Sg segmentation zones. Each of the maps is represented as a set of J characteristic zones , each of the characteristic zones corresponding in the FM map _i to region j of the SgL segmentation. It is possible that certain characteristic areas are empty. The FM _i card values are encoded taking into account SgL segmentation. More precisely, each of the J characteristic zones is coded sequentially, for example in the order of their indexing. Thus, each of the characteristic areas is coded independently of the other characteristic areas. The coded data of each coded area in each of the FM characteristic charts _i is noted Each area undergoes entropic coding. Thus, the coding of the M FM _i cards corresponds to the independent coding of the J zones characteristic of the M FM _i cards.

In a first implementation example, the FMC module only encodes the values of points in the characteristic zone ZoL ^j , excluding any other points in the characteristic maps FM _i . Thus, the encoding of a characteristic zone includes the successive coding for each FM card _i of the values of the points in the characteristic zone in the FM _i card.

Alternatively, the FMC module creates a secondary feature map. for each characteristic zone ZoL ^j . The value of a point on the secondary characteristic map being equal to the value of point _i on the FM chart if that point belongs to characteristic zone ZoL ^j , and to a predetermined value, for example zero, otherwise. The FMC module then encodes the entire characteristic chart to obtain the encoded data different for each characteristic zone ZoL ^j .

As is known, the coding simulation can be identical to the actual coding, or it can approximate it. The coding module quantifies, if necessary, the latent representation of the values in the first group of maps using a quantifier to generate an ordered collection of quantized values. The coding module then compresses the quantized data, using a coding method that takes into account the neighborhood of a value to be coded in the feature map.

The EVAL module performs an evaluation and minimization of coding performance. The evaluation function is, for example, of the rate-distortion type. Minimization can be performed by gradient descent, or any other method within the grasp of a person skilled in the art.

The MAJ module performs an update of the values of the FM _i cards and/or the parameters of the neural network to be encoded, according to the results of the performance function.

There FIG. 2 schematically represents a decoding device DEC of a zone to be decoded Zod ^j of a signal, called zone to be decoded, said zone to be decoded Zod ^j comprising a plurality of samples Pd _n to be decoded.

This DEC decoding device includes an NND module for decoding neural network(s) capable of decoding the NNSYN’ synthesis neural network, a SEGD module for decoding a segmentation, an FMD module for decoding feature maps, an XTR’ module for data extraction, an SE’ module for inverse transformation, and a TT’ module for inverse processing and quantization.

The DEC decoding device produces at output a decoded image including at least the decoded area, denoted Zod ^j (Pd _n ), comprising a plurality of decoded samples Pd _n .

The DEC decoding device in Figure 2 receives as input the Sgc encoded segmentation and a group of encoded data .

The DEC decoding device of the FIG. 2 also receives as input the Wc _k encoded parameters of the NNSYN' synthesis neural network.

The parameters of the NNSYN' synthesis neural network decoded by the NND module are noted Wd _k .

The SEGD module decodes the MZo ^j mask from the Sgc encoded data.

• La carte de segmentation correspondant aux J masques MZo^jest décodée à partir des données Sgc et le masque MZo^jest extrait de la carte de segmentation.
• Le masque MZo^jest décodé directement lorsqu’il a été précédemment encodé dans les données encodées Sgc indépendamment des autres masques.

According to embodiments as described for the encoder:

• The segmentation map corresponding to the J MZo masks ^j is decoded from the Sgc data and the MZo mask ^j is extracted from the segmentation map.
• The MZo mask ^j is decoded directly when it has been previously encoded in the Sgc-encoded data independently of other masks.

The FMD module builds M decoded maps using the MZo ^j mask and encoded data Depending on how the data is encoded Implemented by the ENC encoder and using the MZo ^j mask in the signal domain to be decoded, the FMD module determines the corresponding ZoL ^j characteristic area in the latent domain, and then the FMD module decodes the data to obtain the point values for this characteristic zone ZoL ^j in each of the maps .

In one embodiment, the SE' module performs a transformation of the first group of feature cards decoded to generate a second set of feature maps at the same resolution as the signal to be decoded, noted or The SE' module optionally performs inverse quantization, corresponding to the quantization performed by the encoder. Inverse quantization is not necessary if the encoder's quantizer Q simply rounded the input values. Inverse quantization is also unnecessary if the neural network is capable of handling quantization of its input data. Otherwise, the decoder performs the inverse operation of quantizer Q. The SE' module then transforms the feature map values, including, for example, oversampling, interpolation, filtering, etc., similar to the transformation performed by the encoder. After the transformation, a transformed feature map from the second group has the same resolution as the images in the sequence to be decoded.

In one embodiment, the SE’ module is absent; in this case, the values that will be used to construct the characteristic vector are extracted from the first group of characteristic maps.

The XTR' module is identical to the XTR module in Figure 1. It performs value extraction from the M feature maps. (Or according to one of the embodiments described above), for a sample Pd _n to be decoded, as a function of its coordinates in the signal to be decoded and of the mask MZo ^j .

In one embodiment, the extracted values constitute the vector Zd _n . Zd _n is an L-tuple, that is, it contains L elements, or data zd _i.

In one embodiment, the optional TT' module processes the extracted values to generate the vector Zd _n . The TT' module can perform inverse quantization of the data extracted from the feature set. The processing may include other operations, such as filtering, scaling, etc., similar to those performed by the encoder.

The NNSYN' module is a synthetic neural network, defined by K parameters Wd _k , capable of processing the input vector Zd _n , or L-tuple, to generate as output a second vector representing the sample Pd _n to be decoded, generally a vector containing A elements. In one embodiment, K=3 and the output vector is the triplet (R, G, B) of the decoded pixel Pd _n . The NNSYN' module has the same structure as the NNSYN module, and its parameters are either identical if the encoding of its parameters W _k , or different if the encoding is lossy.

When all the samples Pd_nsignal have been decoded, we have a reconstructed Zod signal^j(Pd_n) of the Zo zone^j.

Furthermore, the DEC decoding device can successively be implemented to decode all the Zod ^j areas in order to reconstruct all the samples to be decoded of the signal, i.e. of the image I(Pd _n ).

The DEC decoding device can be implemented using an electronic device comprising a processor and memory, not shown. Each of the modules mentioned above can then be implemented through the cooperation of the processor and computer program instructions stored in the aforementioned memory and designed to perform the functionalities of the module in question, particularly as described below, when these instructions are executed by the processor. The DEC device can also include multiple processors, with each processor dedicated to the parallel decoding of image areas.

There FIG. 3 illustrates an example of a synthetic artificial neural network used for encoding and decoding in the context of embodiments of the invention.

The synthetic artificial neural network used for encoding, NNSYN, and the synthetic artificial neural network used for decoding, NNSYN’, are defined by an identical structure, including for example a plurality of layers of artificial neurons, and by a set of weights and activation functions associated respectively with the artificial neurons of the network concerned.

The synthetic neural network is, according to one embodiment, an MLP, or Multi Layer Perceptron, followed by one or more layers of convolutional neurons _C1 , … _Cn , each of the layers of convolutional neurons being associated with a filtering mask of predefined size, for example of size 3x3.

MLP consists of an input layer adapted to the input format (the L-tuple), optionally one or more hidden layer(s), and an output layer providing an intermediate output vector Vs _n , generally a vector containing A' elements.

Thus, a vector representation of a current sample (a vector _Zn or _Zdn from the _FMi / _FMSi feature maps to the encoder or / to the decoder) is applied to the input (i.e., to the input layer) of the MLP which produces at the output the intermediate output vector Vs _n .

The concatenation of all these intermediate output vectors constitutes intermediate latent value maps which are then successively processed by the convolutional neuron layer(s) to provide as output a set of output vectors also containing A elements.

According to one embodiment, A is equal to 3, the intermediate output vector is a triplet and the output vector is the triplet (R, G, B) of the pixel P' _n encoded then decoded.

The concatenation of all these triplets in an image constitutes the reconstructed signal I(Pd _n ), according to an example an image I, or Zod ^j (Pd _n ), according to an example the decoded area Zod ^j .

At the encoder, the NNSYN synthetic artificial neural network is trained on the image, minimizing the differences between the input representation of the current image I(P _n ) and its output representation I(P' _n ), while also minimizing the amount of data to be encoded. The EVAL module performs a performance measurement in this regard.

Once the encoder training is complete, the network parameters are encoded, either losslessly, in which case the NNSYN’ neural network is identical to NNSYN, or lossily, in which case the NNSYN’ network may be slightly different from NNSYN.

With reference to the FIG. 4 and to the FIG. 5 , we will now present the application of a convolutional neural layer C _k of the synthesis neural network NNSYN/NNSYN' to a group of input latent value maps CCE ₁ , CCE ₂ , CCE ₃ , to obtain at the output of the convolutional neural layer C _k a group of output latent value maps CCS ₁ , CCS ₂ , CCS ₃ .

In the example described here, the convolutional neural layer C _k is associated with a convolutional kernel of predefined size, for example of size 3x3. This convolutional neural layer C _k transforms three input latent value maps CCE ₁ , CCE ₂ , CCE ₃ into three output latent value maps CCS ₁ , CCS ₂ , CCS ₃ .

The application of this convolutional neural layer C _k to the three input latent value maps CCE ₁ , CCE ₂ , CCE ₃ includes, in a step (1), obtaining a characteristic region R ^j comprising all the points of the maps CCE ₁ , CCE ₂ , CCE ₃ colocalized with samples belonging to the mask MZo ^j of the Zod ^j area. The points of this characteristic region R ^j are called determined points.

Then, during a step (2), for a point CCS ₁ , CCS ₂ , CCS ₃ maps colocalized with a sample Pt _n belonging to the mask MZo ^j of the area to be decoded Zod ^j , an input vector Vco _n of the convolutional neural layer C _k is constructed.

This input vector Vco _n depends on the point Pa _n of the CCE ₁ , CCE ₂ , CCE ₃ maps co-located with the point CCS ₁ , CCS ₂ , and CCS ₃ maps, along with contextual information about point Pa _n, are used. This contextual information is represented by a neighborhood of point Pa _n , which by definition includes point Pa _n itself. In the example presented here, the neighborhood used comprises 9 points associated with a 3x3 square mask centered on point Pa _n .

This contextual information is also conditioned by the feature region Rj in the input maps CCE1 , CCE2 , and CCE 3 . More precisely, the contextual information depends on the location of point Pn relative to the boundary of area Rj . This location is obtained by evaluating a proximity criterion Cp(Pn ) of point Pn relative to the boundary of feature region Rj . In the example described here, the proximity criterion Cp(Pn  ) is the Euclidean distance between point P n and the boundary of feature region R j . Alternatively, the proximity criterion Cp(Pn ) could be a Manhattan, Minkowski, or Chebyshev distance. In another variant, the distance can simply be measured in pixels, for example, by counting the number of pixels along the two directions of a coordinate system associated with the current map. In yet another variant, the proximity criterion Cp(Pa n ) may also be a quasi-distance or a gap or any other relevant measure indicating the proximity of the point Pa n to the boundary of the characteristic region R j .

• Dans le premier cas, illustré à laFIG. 4, le critère de proximité Cp(Pa_n) est supérieur (éventuellement égal) à une distance d (par exemple prédéfinie) et les valeurs de voisinage des cartes d’entrées CCE₁, CCE₂, CCE₃ sont utilisées pour former le vecteur d’entrée Vco_n.
• Dans le deuxième cas, illustré à laFIG. 5, le critère de proximité Cp(Pa_n) est inférieur à cette distance d (dans l’exemple décrit ici, le point Pa_nest localisé à une distance nulle de la frontière de la région caractéristique R^j). Dans ce cas, le vecteur d’entrée Vco_nest défini composante par composante. La valeur d’une composante du vecteur d’entrée Vco_nest égale à la valeur du point associé de la carte d’entrée si ce point appartient à la région caractéristique R^jet à une valeur de remplacement sinon.

Two cases must then be distinguished:

• In the first case, illustrated in theFIG. 4, the proximity criterion Cp(Pa_n) is greater than (or possibly equal to) a distance d (for example, predefined) and the neighborhood values of the CCE input maps₁, CCE₂, CCE₃ are used to form the input vector Vco_n.
• In the second case, illustrated in the FIG. 5 The proximity criterion Cp(Pa _n ) is less than this distance d (in the example described here, point Pa _n is located at zero distance from the boundary of characteristic region R ^j ). In this case, the input vector Vco _n is defined component by component. The value of a component of the input vector Vco _n is equal to the value of the associated point on the input map if that point belongs to characteristic region R ^j , and to a replacement value otherwise.

In step (3), the input vector is processed by the convolutional neural layer C _k to generate the point as output CCS ₁ , CCS ₂ , CCS _{3 cards.}

In the FIG. 4 and the FIG. 5 The values shown in gray are part of the characteristic region Rj and are used to determine components of the input vector Vco n . The values shown in white ( FIG. 5 ) do not belong to the characteristic region R j but to another region R x and cannot be used to define the input vector Vco n , they are replaced by a replacement value Rp.

In one embodiment, the replacement value is a predetermined constant value, for example equal to 0.

In another embodiment, the replacement value is a function of the values of the neighborhood points belonging to the characteristic region R_^j; for example, the replacement value is equal to the value of point P_{_n}. In yet another embodiment, a set of replacement values is calculated based on the values of the neighborhood points belonging to the characteristic region R_^j. A neural network can be implemented to calculate a given number of replacement values based on the neighborhood points of the characteristic region R_^j.

In the preceding embodiments, the definition of the neighborhood of point Pa _n is independent of the proximity criterion Cp(Pa _n ). In another embodiment, and as represented in the FIG. 6 , the definition of the neighborhood of the point Pa _n can depend on the proximity criterion Cp(Pa _n ).

• Dans le premier cas, le critère de proximité Cp(Pa_n) est supérieur (éventuellement égal) à la distance d et les valeurs associé à un premier voisinage du point Pa_n(par exemple identique à celui défini précédemment en lien avec laFIG. 4) des cartes d’entrées CCE₁, CCE₂, CCE₃ sont utilisées pour former le vecteur d’entrée Vco_n.
• Dans le deuxième cas, le critère de proximité Cp(Pa_n) est inférieur à cette distance d. Dans ce cas un deuxième voisinage du point Pa_nest utilisé. De façon préférentielle, pour respecter la topologie de la couche neuronale convolutive C_k, le nombre de points du premier voisinage est égal au nombre de points du deuxième voisinage. Dans l’exemple présenté à laFIG. 6, le deuxième voisinage est par exemple obtenu à partir du premier voisinage en adaptant la forme de ce dernier à la frontière de la région caractéristique R^jde sorte que les points du deuxième voisinage appartiennent tous à la région caractéristique R^j.

Two cases must again be distinguished:

• In the first case, the proximity criterion Cp(Pa_n) is greater than (possibly equal to) the distance d and the values associated with a first neighborhood of point Pa_n(for example, identical to that defined previously in connection with theFIG. 4) CCE entry cards₁, CCE₂, CCE₃ are used to form the input vector Vco_n.
• In the second case, the proximity criterion Cp(Pa _n ) is less than this distance d. In this case, a second neighborhood of the point Pa _n is used. Preferably, to respect the topology of the convolutional neural layer C _k , the number of points in the first neighborhood is equal to the number of points in the second neighborhood. In the example presented at the FIG. 6 , the second neighborhood is obtained for example from the first neighborhood by adapting the shape of the latter to the boundary of the characteristic region R ^j so that the points of the second neighborhood all belong to the characteristic region R ^j .

Once determined, the input vector is processed, in step (3), by the convolutional neural layer C _k to generate the point as output CCS ₁ , CCS ₂ , CCS ₃ cards _.

There FIG. 7 is a flowchart representing an example of a coding process that can be implemented by the coding device of the FIG. 1 when the NNSYN synthesis neural network is, for example, the one presented at the FIG. 3 .

According to this embodiment, the signal is a two-dimensional image, each sample to be coded is therefore a pixel P _n with coordinates (x _n , y _n ).

The encoding process takes place in three main phases:

In a first phase, called the segmentation phase, the segmentation Sg of the input signal I(P _n ) into J (greater than or equal to 2) zones Zo ^j is carried out.

In a first example of segmentation, the image is divided into J regular zones, for example, of identical size and shape (except possibly at the edges of the image). For example, the image can be divided into rectangular zones with horizontal and vertical boundaries. Such a division corresponds to the concept of "tiles" implemented by encoding standards such as H.264/AVC, H.265/HEVC, and H.266/VVC.

According to a second example of segmentation, the image is divided into J zones corresponding to samples (or blocks) scanned in a lexicographical order. Such a division corresponds to the concept of "slices" implemented by coding standards such as H.264/AVC, H.265/HEVC and H.266/VVC.

In a third segmentation example, the image is divided into semantic zones, such as a background and a foreground. This can be done manually by an operator. Segmentation can also be automatic or semi-automatic, depending on the segmentation algorithm used. It's worth noting that there are no restrictions related to the type of segmentation algorithm used.

Advantageously, such a division into zones allows, during the subsequent decoding of the signal I(Pd _n ) to parallelize the decoding of the zones to be decoded Zod ^j by distributing the decoding load equally on each of the decoders and/or processors.

In a second phase, called the construction phase, a learning process is performed to determine, for an input signal I(Pn ), the values of the FM i maps and the parameters W k  to optimize an overall cost function. This learning is, for example, carried out by gradient descent, followed by updating the parameters of the NNSYN synthesis neural network and the values of the FMi feature maps. As is known in the state of the art, the cost function can be of the rate-distortion type, or rate-distortion type, or perceptual type. To measure the rate R, it is necessary to simulate the encoding of the J ZoLj zones of the maps. and then measure the associated coding rate (the size of the B2 stream). In one embodiment, the coding of the Wk parameters is not simulated because their influence is less significant than that of the feature maps. In another embodiment, the coding of the Wk parameters is also simulated, and the associated rate (the size of the B1 stream) is measured. To measure the distortion D, it is necessary to simulate the coding and then the decoding of at least a portion of the image I, to obtain at least one pixel P'n resulting from a coding and then decoding simulation, and then measure the difference between this portion of the input image I( Pn ) and a corresponding portion of the encoded and then decoded image I( P'n ).

Then, during a third phase, called the coding phase, the Sg segmentation, the data from each zone and the W _k parameters are encoded to produce the Sgc-encoded values, and Wc _k before transmission or storage. They constitute the compressed representation of the input signal I(P _n ), this compressed signal being able to be decoded zone by zone.

We will now describe the steps of a coding process according to one embodiment of the invention.

During an E20 step, a signal I(P _n ) to be coded, comprising a plurality of N samples P _n , is provided as input to the process.

During an E21 step, the segmentation Sg of the input signal I (P _n ) into J (greater than or equal to 2) zones Zo ^j is carried out.

During step E22, the M FM _i maps of the first group are initialized. Subsequently, the parameters W _k of the NNSYN synthesis neural network and the values of the FM _i maps must be optimized during the construction phase.

According to one embodiment, the FM _i cards have the same resolution as the input signal I (P _n ) and therefore each have the same number of values N as there are samples P _n to be coded.

According to one embodiment, the FM _i cards have a resolution less than or equal to that of the input signal I(P _n ) and therefore include, for at least one of them, a number N' of values to be coded less than N. According to a variant, the first FM _i card has the resolution of the images and each subsequent card has half the resolution of the previous one.

According to one embodiment, several FM _i cards have the same resolution, lower than that of the input signal I(P _n ).

In one embodiment, the FM _i maps are transformed to provide a second group of transformed feature maps, FMS _i . In this embodiment, the feature vectors are preferably extracted from the transformed maps of the second group, and not directly from the maps of the first group. Thus, in this embodiment, the feature vectors are indirectly extracted from the maps of the first group. The maps of the second group are not coded; they are used solely for constructing the feature vectors.

According to one embodiment, the FM _i cards are initialized with predefined constant values.

According to another embodiment, the feature maps are initialized by a set of random real values.

The FM _i feature maps of the first group are subsequently updated, or refined, during an E23 step , by the encoder update module during its learning.

During an E24 stage, the J zones FM _i cards from the first group are encoded by the FMC module of the encoder. During the construction phase, this operation is a coding simulation. During the encoding phase, this operation is the actual encoding, and the encoded values constitute the B2 stream. The simulation can be identical to the actual encoding, but it can also be different (for example, simplified). For this encoding, a technique for predicting a feature card value based on its neighborhood is used, as will be described, for example, in support of the FIG. 11 .

In one embodiment, the J zone ZoL ^j of the FM _i cards are coded in the indexing order of the J zones and in a predefined order of the associated feature cards (i.e., …) and values for each map, for example lexicographic. Each zone ZoL ^j of the FM _i maps undergoes entropic coding.

The entropic coding of all zones of the set of FM cards _i produces a compressed stream B2 whose throughput is measured later during a step E29.

In an E25 step, according to one embodiment, the M cards of the first group FM _i are transformed by the SE module to generate cards of the second group FMS _i at the resolution of the images of the input sequence.

According to one embodiment, M FMS _i cards are generated.

According to one embodiment, each FM _i card is transformed into an FMS _i card.

In one embodiment, at least one _FM1 map has a lower resolution than the images in the sequence to be encoded, and the transformation operation includes oversampling so that the transformed FM1 map has the same number of samples as the images in the sequence. Oversampling consists of adding values to the _FM1 maps to achieve the resolution of the input sequence images. It can be simple (by nearest neighbor replication) or involve interpolation (linear, polynomial, filtered, etc.).

During step E26 and taking into account the segmentation Sg of the input signal I(P_n) in J zones Zo^jValues are extracted by the XTR module from the FM cards_ior possibly FMS_itransformed_.This extraction is performed based on the coordinates (x_n, y_n) of sample P_nof the input signal and possibly the MZo mask^jIt can also be done according to the resolution of the map in question.

According to one embodiment, the characteristic vector Z _n results directly from this extraction.

The samples to be coded are, for example, processed in sequential order, from n=1 to n=N.

According to one embodiment, during step E27, the vector Z_nThe characteristic is constructed by the TT module from the values extracted from the FM cards._ior FMS_ifor each sample P_ncoordinates (x_n, y_n) of the input signal. The processing may include quantifying the values extracted from the FM cards_ior the vector Z_nconstituted, if necessary. The processing may include other operations, such as filtering, scaling, the application of any function, preferably monotonic, etc.

In one embodiment, Z_nis an M-tuple (z₁, z₂,…, z_J), consisting of the FM card values_ior FMS_ilocated at coordinates (x_n, y_n) of the current pixel P_n, as will be illustrated in support of theFIG. 8.

In one embodiment, Z _{_n} is an M-tuple constructed from values taken from the FM_{_i} maps at coordinates that may differ between maps. For example, if the FM_{_i} maps have different resolutions because they have been downsampled, the coordinates are adjusted (by scaling) to match the resolution of each map.

During an E28 step, the vector Z _n is processed by the NNSYN synthesis neural network to generate as output a vector representative of the sample P _n to be coded, according to an embodiment the triplet (R, G, B) of the sample P' _n (the sample P _n coded then decoded).

The structure and parameters _Wk of the synthesis neural network are initialized, for example, during the first iteration of this step. These parameters are subsequently updated, or refined, during the construction phase, in later iterations of the process.

According to one embodiment, the parameters of the synthesis neural network and/or the prediction neural network are initialized by predefined values known to give a satisfactory result (for example, following training on a corpus of images).

According to another embodiment, the parameters of the synthesis neural network and/or the prediction neural network are initialized by a set of random values.

During step E29 , the W_{_k} parameters of the NNSYN synthesis neural network are quantized and encoded. During the construction phase, this operation is a coding simulation. During the encoding phase, this operation is the actual encoding, and the encoded values constitute the B1 stream. The simulation can be identical to the actual encoding, but it can also be different (for example, simplified). Any known technique can be used for this purpose, for example, the neural network coding standard proposed by MPEG-7 Part 17, also called NNR (Neural Network Representation). Note that in this case, the amount of degradation that the encoding introduces to the W _{_k} weights must be chosen.

During an E30 step, a performance measure is evaluated .

For this purpose, the coding simulation rates associated with the feature maps of the first group (simulation of the B2 flow by coding the J zones) are used. FM maps _i ) and optionally the parameters of the neural network(s) (simulation of the B1 flow by encoding the parameters W _k ) are measured.

According to one embodiment, the cost function is of type distortion-rate, denoted (D+L*R), where D, for example, is the squared error measured between the input signal and the decoded signal (or the error measured on a subset of the signal samples). In another example, D is calculated from a perceptual function such as SSIM (forStructural SIMILarity), or MSSSIM (forMulti-scale Structural SIMilarityIn one embodiment, R is the simulated bit rate of stream B1; in another embodiment, R is the total bit rate used to encode this image, i.e., the sum of the simulated bit rates of B1 and B2. L is a parameter that controls the bit rate-distortion trade-off. Other cost functions are possible.

As long as the cost function has not reached its minimum, or a maximum number of iterations of the cost function minimization algorithm has been reached, the performance measurement is not satisfactory, and the process is repeated from step E23. This minimization can be performed by a mechanism known as gradient descent with parameter updates during step E23 for the feature map values and E29 for the network(s) parameters.

During a finite element (FE) step, if the cost function has reached its minimum, or if a maximum number of iterations of the cost function minimization algorithm has been reached, training stops. If a coded version corresponding to the last simulation of the synthetic neural network parameters (W)_k) and characteristic maps ( is available, and streams B1 and B2 can be composed of them. According to another embodiment, the effective encoding of the updated parameters of the synthesis neural network (W_k) and feature map values (FM_i) is performed at this stage to produce the Wc-encoded parameters_kAnd which constitute B1 flows and B2. Furthermore, during this step, the Sg segmentation encoding is performed to produce an encoded segmentation of the J MZoc masks^jwhich constitutes the B3 flow.

• un pointeur de flux (qui indique l’adresse du début de chaque sous-flux dans le flux global), ou
• un marqueur (une série de bits interdite par ailleurs, qui permet de parcourir le flux pour trouver le début de chaque sous-flux), ou
• tout autre moyen permettant d’identifier une sous-partie d’un flux codé.

The B2 stream comprises sub-streams corresponding to each area of the feature maps. When coding each latent zone, it is possible, in a preferred embodiment, to place location information for the sub-streams within the overall stream at an identifiable point in the flow. This location information can be:

• a flow pointer (which indicates the starting address of each subflow in the overall flow), or
• a marker (a series of bits otherwise forbidden, which allows traversing the stream to find the beginning of each sub-stream), or
• any other means of identifying a sub-part of an encoded stream.

The B1 and B2 streams can be multiplexed and/or concatenated to produce a final stream. In one embodiment, the B3 stream of the coded segmentation and the B1 stream of the coded parameters of the neural network(s) are stored or transmitted before the B2 stream, so that they can be decoded before the B2 stream.

It should be noted that only one NNSYN neural network is used for encoding each Zo ^j area.

There FIG. 8 represents in an illustrated manner a coding method used in an embodiment of the invention.

In this embodiment, there are 4 FM _i cards generated. In a preferred mode, there are 7.

The first FM map ₁ has the same resolution as the image I(P _n ), and therefore contains WxH variables, where W represents the image width in pixels, and H its height. The second FM map ₂ has half the resolution (in each dimension) of the FM map _1. Each additional map has half the resolution of the previous map. This structure reduces the number of variables in the feature maps, which facilitates coding and learning while minimizing coding costs.

The FM ₂ card is oversampled by the SE module by a factor of 2 in each dimension, according to a process illustrated in support of the FIG. 6 The FM ₃ map is oversampled by a factor of 4 in each dimension, and the FM ₄ map by a factor of 8 in each dimension. The FM ₁ map is not affected by the oversampling. (FMS ₁ = FM ₁ ).

The resulting FMS _i maps are of the same resolution as the image I(P _n ), and therefore each have WxH values, where W represents the width of the image in pixels and H its height (N=WxH).

Other types of structure are possible, for example one can use a different reduction rate than half between the cards (a quarter, or a third, etc.).

In this embodiment, the vector Z _{_n} is a 4-tuple (z _₁ ...z _₄ ) consisting of the values extracted from the FMS maps _i located at the coordinates (x _{_n} , y _{_n}) of the current pixel P _{_n} . The vector Z _{_n}, consisting of the extracted (quantized) values from the FMS maps _i, is processed by the NNSYN synthesis neural network to generate a second output vector. In this example, the output vector is the triplet (R, G, B) of the encoded and then decoded pixel P'_{_n}. The triplet is inserted into the decoded image I(P'_{_n}) at the positions (x _{_n} , y _{_n} ) of the color components (R', G', B').

In another embodiment, not shown, the vector Z _n is extracted directly from the FM _i layers, at positions recalculated according to the size of the maps, then the extracted values are possibly processed and quantified after extraction.

There FIG. 9 is a flowchart representing an example of a decoding process for a decoded zone Zod ^j that can be implemented by the DEC decoding device of the FIG. 2 when the NNSYN' synthetic neural network is, for example, the one presented at the FIG. 3 .

During step F20, stream B1, a portion of stream B2 (corresponding to the feature maps of zone Zod ^j ), and stream B3 are extracted from the encoded stream. They contain, respectively, Wc _k parameters, an Sgc segmentation, and encoded representations. M maps of the first group representative of the area to be decoded Zod ^j .

During an F21 step, an MZo ^j segmentation mask is generated by decoding the values of the encoded segmentation.

During stage F22, the M cards are generated by decoding the values In one embodiment, the cards are decoded in order ( , ,… ), and the values of each card in a predefined order, for example lexicographical, possibly taking into account (depending on the coding technique implemented by the ENC encoder) the MZo ^j mask.

• Les cartes sont de même résolution que le signal I(Pd_n) à reconstituer, c’est-à-dire qu’elles comportent N=WxH valeurs.
• Les cartes sont à une résolution inférieure ou égale à celle du signal I(Pd_n) à reconstituer.
• Plusieurs cartes sont de même résolution, inférieure à la résolution du signal.

According to embodiments as described for the encoder:

• The maps are of the same resolution as the signal I(Pd _n ) to be reconstructed, that is to say they have N=WxH values.
• The maps are at a resolution less than or equal to that of the signal I(Pd _n ) to be reconstructed.
• Several cards are of the same resolution, lower than the resolution of the signal.

During step F23, according to one embodiment, the M cards of the first group are transformed by the SE' module to generate maps of the second group to the resolution of the input images. This step is similar to step E25, which was described for the encoder supporting the FIG. 5 , and the embodiments apply. In particular:

According to one embodiment, M cards are generated.

According to one embodiment, each card is transformed into a map .

According to one embodiment, at least one map is of lower resolution than the images of the image to be encoded, and the transformation operation includes oversampling so that the map The transformed image contains the same number of samples as the input image. Oversampling involves adding values to the maps to achieve the resolution of the input image. It can be simple (by nearest neighbor replication) or involve interpolation (linear, polynomial, by filtering, etc.).

The transformation may optionally include inverse quantization of the extracted values, if necessary. However, inverse quantization is not mandatory.

During an F24 stage, values are extracted by the XTR module from the cards or possibly transformed_.This extraction is performed based on the coordinates (x_n, y_n) of a sample P_nof the input signal. It can also be performed depending on the resolution of the card in question. This step is similar to step E26, which was described for the encoder supporting theFIG. 7, and the embodiments apply.

In one embodiment, Zd_nis an M-tuple (z₁, z₂,…, z_J), consisting of the card values Or located at coordinates (x_n, y_n) of a current pixel Pd_n, as will be illustrated in support of theFIG. 10.

The samples to be decoded are, for example, processed in sequential order relative to the MZo ^j mask.

According to one embodiment, during step F35, a vector Zd _n is constructed by the module TT' from the values extracted from the maps from the first group or the cards of the second group, for each sample Pd _n of coordinates (x _n , y _n ) to be decoded, as a function of the coordinates (x _n , y _n ). This step is similar to step E27 which was described for the encoder supporting the FIG. 7 and the described embodiments apply. The extraction may include inverse quantization of the extracted values or of the resulting vector Zd _n , if necessary.

During step F26 , the Wd _k parameters of the NNSYN' synthesis neural network are generated by decoding the Wc _k values of the B1 stream. Any known decoding technique corresponding to the encoding technique used by the encoder can be used for this purpose. The NNSYN' synthesis neural network is similar to the NNSYN synthesis network; that is, it has the same structure and parameters, except for the encoding, which can be performed with or without loss.

In one embodiment, stream B1 is decoded before streams B2 and B3, so that the NNSYN’ synthesis neural network is available before decoding the samples. Similarly, stream B3 is decoded before stream B2 so that the mask of the region to be decoded is available before decoding the samples.

During step F27, the vector Zd _n is processed by the NNSYN' synthesis neural network to generate a second vector representing the sample Pd _n to be decoded, in one embodiment a triplet which is injected into the image of the decoded area Zod ^j (Pd _n ) at the positions (x _n , y _n ) of the color components (Rd, Gd , Bd ). This step is similar to step E28 which was described for the encoder supporting the FIG. 5 .

When all the signal samples have been processed, the decoded signal corresponding for example to the Zod ^j (Pd _n ) zone is available.

It should be noted that only one neural network NNSYN' is used during decoding regardless of the Zod ^j area being decoded.

There FIG. 10 represents in an illustrated manner a method of decoding a Zod ^j zone used in an embodiment of the invention.

In this embodiment, the cards are decoded There are 4 of them. In a preferred mode, there are 7.

In this embodiment, the cards are decoded are representative only of the Zod ^j zone. In other words, only the data from the ZoL ^j zone (corresponding in the latent domain to the Zod ^j zone) have been decoded.

In this embodiment, the first card has the same resolution as image I, and therefore has variable WxH, where W represents the width of the image in pixels, and H its height. The second map is half the resolution (in each dimension) of the map Each additional map has half the resolution of the previous map. This structure reduces the number of variables in the feature maps, which facilitates decoding while minimizing coding costs.

The map is oversampled by a factor of 2 in each dimension, using any oversampling method available to a person skilled in the art. The map is oversampled by a factor of 4 in each dimension, and the map by a factor of 8 in each dimension.

The maps are of the same resolution as the image to be decoded, and therefore include WxH values, where W represents the width of the image in pixels, and H its height.

In this embodiment, the vector Zd _n is a 4-tuple (z ₁ ...z ₄ ) consisting of the values of the maps located at the coordinates (x _{_n}, y_{_n} ) of the current pixel Pd _{_n} . The vector Zd _{_n} is optionally dequantized and then processed by the NNSYN' synthesis neural network to generate the (R, G, B) triplet representative of the sample Pd_{_n} to be decoded. The (R, G, B) triplet is inserted into the decoded image I(Pd _{_n} ) at the coordinates (x _{_n}, y _{_n} ) in the color components (Rd, Gd, Bd).

There FIG. 11 is a flowchart representing a method for encoding feature cards that can be implemented by the encoding device of the FIG. 1 and by the coding process of the FIG. 5 .

These steps are substeps of step E30 described previously, supported by Figure 7. Their purpose is to encode a current value V_nfrom a point in a characteristic area of a characteristics map of the first group currently being processed using neighborhood values.

During substep E301, a neighborhood vector C_nis established, including values close to the value V_nThese neighboring values can be located in the same map and/or in a different map from the plurality M of maps _.This neighborhood vector consists of of a number C of values, or data, corresponding to neighborhood values (for example, C=10). These values must be known to the encoder and the decoder; therefore, they must be in a causal neighborhood of the value V_n.

According to a first embodiment, these values are used to determine the context of an entropy encoder to encode the current value during an E303 step. This encoder can be a CABAC (Context-Adaptive Binary Arithmetic Coding) encoder. This type of encoder is well known to those skilled in the art. It is notably used in the H.265/HEVC video compression standard. It is an arithmetic encoder with lossless compression. It decomposes all non-binary symbols into binary symbols. Then, for each bit, the encoder selects the most suitable probability model and uses a context to optimize the probability estimation. This context can be defined by information from neighboring elements. Arithmetic coding is then applied to compress the resulting data. As is known to those skilled in the art, there are several ways to use the neighborhood vector to produce context information. For example, one can count the number of neighboring values that are not zero and associate a context with each number. Alternatively, one can perform comparisons between several neighboring values and associate a given context with a configuration of order between the neighboring values, for example by ranking the neighboring values in ascending order and associating a context with each possible order.

In a second embodiment, the neighborhood is used to predict, during step E302 , the current value from an autoregressive model. Recall that an autoregressive model predicts a sample from a series based on its past values. In this embodiment, the past values are constituted by the context, and the difference between the predicted variable and the actual value is quantified and then entropically coded during step E303.

At the end of the process, the current coded value Vc_nof the map currently being processed is coded.

There FIG. 12 is a flowchart representing a method for decoding feature cards that can be implemented by the decoding device of the FIG. 2 and by the decoding process of the FIG. 7

These steps are substeps of step F22 described previously in Figure 7. Their purpose is to decode a current value of a point in a characteristic area of a characteristics map of the first group currently being processed using neighborhood values.

During a substep F221, a neighborhood vector Cd_nis established, including values close to the value Vd_nThis step is similar to step E301 described previously and the same embodiments apply. This neighborhood vector consists of of a number C of values, or data, corresponding to neighborhood values (for example, C=10) located in the same map and/or in a different map of the plurality M of maps of a point in a characteristic area of a characteristics map These values are located in a causal neighborhood of the value Vd_nare known to the decoder.

According to a first embodiment, these values are used to determine the context of an entropy decoder for decoding the current value during an F223 step. This decoding is similar to that used in the encoder, for example, CABAC. The use of the neighborhood to produce context information is similar to that chosen for the encoder. For example, one can count the number of non-zero neighboring values and associate a context with each number. Alternatively, one can perform comparisons between several neighboring values and associate a given context with an ordering configuration among the neighboring values, for example, by sorting the neighboring values in ascending order and associating a context with each possible order.

In a second embodiment, the neighborhood is used to predict the current value from an autoregressive model during step F222. In this mode, the past values are constituted by the context, and the difference between the predicted variable and the actual value is quantified and then entropically coded during step F223.

At the end of the process, the current decoded value Vd _n of the FMd _i card being processed is decoded.

It should also be noted that the invention is not limited to the embodiments described above. Indeed, it will be apparent to those skilled in the art that various modifications can be made to the embodiments described above, in light of the information just disclosed to them.

For example, NNSYN/NNSYN’ synthetic neural networks can be recurrent neural networks.

In another example, synthetic neural networks can consist of one or more convolutional neural networks, followed by a multi-level programming (MLP) network, and then followed by one or more convolutional neural networks, each convolutional neural network being associated with a convolutional kernel of predefined size, for example, 3x3. In these examples, obtaining the vectors Z _{_n} /Z_d is adapted to the topology of the synthetic neural networks NNSYN/NNSYN' in a similar way to adapting the input vector V_co _n shown with reference to Figures 4 to 6.

In the detailed presentation of the invention given above, the terms used shall not be interpreted as limiting the invention to the embodiments set forth in this description, but shall be interpreted as including all equivalents which can be foreseen by a person skilled in the art by applying their general knowledge to the implementation of the teaching which has just been disclosed to them.

Claims (14)

Method for coding a zone (Zo^j) of a signal (I(P_n)), said coding area, said coding area comprising a plurality of samples (P_n) to be coded, said coding process comprising the following steps:

• obtaining a first group of at least one latent value map (FM _i ) representative of said signal (I(P _n )),
• obtaining a mask (MZo ^j ) in said signal (I(P ⁿ )) of said zone to be coded (Zo ^j ),
• for at least one sample of said zone to be coded (Zo ^j ), called current sample (P _n ), associated with a position (x _n , y _n ) in said signal (I(P _n )) to be coded:
  • the construction of a characteristic vector (Z _n ) from said latent value maps (FM _i ) of said first group, as a function of said position (x _n , y _n ) of said current sample (P _n ),
  • the processing of said characteristic vector (Z _n ) by an artificial neural network, called a synthetic neural network (NNSYN), said synthetic neural network being defined by a set of parameters (Wk) and comprising at least one synthetic neural layer (c _k ) to obtain, at the output of said synthetic neural network, a vector (P' _n ) representative of a decoded value of the current sample (P _n ),
  • the updating of at least one value of one of said latent value maps of said first group and/or of at least one parameter of said synthetic neural network, as a function of a coding performance measure,
• obtaining a second group of latent value maps representing at least said zone to be coded (Zo ^j ) from said mask (MZo ^j ),
• the encoding of the second group of latent value maps ( ),
• the encoding of said mask (MZo ^j ), and
• the coding of said set of parameters (W _k ) of said synthetic neural network.

said processing comprising the application of said at least one neural synthesis layer to transform at least one group of latent input maps (LIMs), referred to as input maps, into at least one group of latent output maps (LIMs), referred to as output maps, said application comprising:

• obtaining a characteristic region (R ^j ) of said zone to be decoded in said input maps from said mask (MZo ^j ), said characteristic region (R ^j ) comprising the points associated with said mask,
• for at least one point (Pt _n ) of said area to be decoded, called point to be decoded:
  • obtaining an associated point (Pa _n ) to the said point to be decoded (Pt _n ) in at least one of the latent input maps (CCE),
  • the evaluation of a proximity criterion of said associated point (Pa _n ) with respect to a boundary of said characteristic region,
  • the construction of an input vector (Vco _n ) of said neural synthesis layer from a neighborhood of said associated point (Pa _n ) and said evaluation of said proximity criterion, and
  • the processing of said input vector by said neural synthesis layer to obtain a point (Ps _n ) of at least one of the output maps,

Method for decoding a zone (Zod)^j), called the decoding area, of a signal comprising at least two areas, said decoding area comprising a plurality of samples (Pd_n) to decode, said decoding process comprising the following steps:

• the decoding (F22) of a group of latent value maps ( ) representative of said zone to be decoded (Zod ^j ) of said signal,
• obtaining (F20) a mask (MZo ^j ) of said zone to be decoded in said signal (I(P _n )),
• the decoding (F26) of a set of parameters (Wd _k ) representative of a neural network (NNSYN'), called a synthetic neural network comprising at least one synthetic neural layer,

• the processing (F27) of said group of decoded latent maps ( ) by said synthetic neural network (NNSYN') to produce at the output of said synthetic neural network at least said decoding area (Zod ^j ), said processing comprising the application of said at least one synthetic neural layer to transform at least one group of input latent maps (CCE), called input maps, into at least one group of output latent maps (CCS), called output maps, said application comprising obtaining a characteristic region (R ^j ) of said decoding area in said input maps from said mask (MZo ^j ), said characteristic region (R ^j ) comprising the points associated with said decoding area (Zod ^j ) and for at least one point (Pt _n ) of said decoding area, called decoding point, the processing of an associated point (Pa _n ) of said decoding point in at least one of the input latent maps (CCE) by said neural layer as a function of its proximity to a boundary of said characteristic region.

Method for decoding a zone (Zod)^j) of a signal according to the preceding claim in which the processing step of said associated point (Pa_n) includes:

• obtaining an associated point (Pan) to be decoded in at least one of the latent input maps (CCE),
• the evaluation of a proximity criterion of said associated point (Pan) with respect to a boundary of said characteristic region,
• the construction of an input vector (Vcon) of said neural synthesis layer from a neighborhood of said associated point (Pan) and said evaluation of said proximity criterion, and
• the processing of said input vector by said neural synthesis layer to obtain a point (Psn) of at least one of the output maps.

Method of decoding an area (Zod ^j ) of a signal according to the preceding claim in which said neighborhood of said associated point (Pa _n ) is independent of said evaluation of the proximity criterion. Method of decoding an area (Zod ^j ) of a signal according to claim 3 in which said neighborhood of said associated point (Pa _n ) is selected according to said evaluation of the proximity criterion. Method of decoding a zone (Zod ^j ) of a signal according to any one of claims 3 to 5 wherein said neighborhood of said associated point (Pa _n ) is selected according to said associated point (Pa _n ). Method of decoding a zone (Zod ^j ) of a signal according to any one of claims 3 to 6 wherein during the construction of the input vector (Vco _n ), a component of the input vector (Vco _n ) is associated with a point of said neighborhood, the value of said component being equal to the value of said associated point if said associated point is a point of said characteristic region (R ^j ) and to a replacement value otherwise. Method of decoding a zone (Zod ^j ) of a signal according to the preceding claim wherein said replacement value is dependent on the points of said characteristic region (R ^j ) or of the points associated with said neighborhood belonging to said characteristic region (R ^j ). Decoding method according to claim 7 wherein said replacement value does not depend on the points of said characteristic region (R ^j ). Method for decoding an area (Zod ^j ) of a signal according to any one of claims 3 to 9 wherein said proximity criterion is a distance. Method for decoding a zone (Zod ^j ) of a signal according to any one of claims 2 to 9 wherein said neural synthesis layer is a convolutional neural layer. Encoding device (ENC) of a signal (I(P_n)) comprising a plurality of samples (P_n) to be coded, characterized in that said coding device is configured to implement:

• obtaining a first group of at least one latent value map (FM _i ) representative of said signal (I(P _n )),
• obtaining a mask (MZo ^j ) in said signal (I(P ⁿ )) of said zone to be coded (Zo ^j ),
• for at least one sample of said zone to be coded (Zo ^j ), called current sample (P _n ), associated with a position (x _n , y _n ) in said signal (I(P _n )) to be coded:
  • the construction of a characteristic vector (Z _n ) from said latent value maps (FM _i ) of said first group, as a function of said position (x _n , y _n ) of said current sample (P _n ),
  • the processing of said characteristic vector (Z _n ) by an artificial neural network, called a synthetic neural network (NNSYN), said synthetic neural network being defined by a set of parameters (Wk) and comprising at least one synthetic neural layer (c _k ) to obtain, at the output of said synthetic neural network, a vector (P' _n ) representative of a decoded value of the current sample (P _n ),
  • the updating of at least one value of one of said latent value maps of said first group and/or of at least one parameter of said synthetic neural network, as a function of a coding performance measure,
• obtaining a second group of latent value maps representing at least said zone to be coded (Zo ^j ) from said mask (MZo ^j ),
• the encoding of the second group of latent value maps ( ),
• the encoding of said mask (MZo ^j ), and
• the coding of said set of parameters (W _k ) of said synthetic neural network.

said processing comprising the application of said at least one neural synthesis layer to transform at least one group of latent input maps (LIMs), referred to as input maps, into at least one group of latent output maps (LIMs), referred to as output maps, said application comprising:

• obtaining a characteristic region (R ^j ) of said zone to be decoded in said input maps from said mask (MZo ^j ), said characteristic region (R ^j ) comprising the points associated with said mask,
• for at least one point (Pt _n ) of said area to be decoded, called point to be decoded:
  • obtaining an associated point (Pa _n ) to the said point to be decoded (Pt _n ) in at least one of the latent input maps (CCE),
  • the evaluation of a proximity criterion of said associated point (Pa _n ) with respect to a boundary of said characteristic region,
  • the construction of an input vector (Vco _n ) of said neural synthesis layer from a neighborhood of said associated point (Pa _n ) and said evaluation of said proximity criterion, and
  • the processing of said input vector by said neural synthesis layer to obtain a point (Ps _n ) of at least one of the output maps,

Zone decoding device (DEC) (Zod^j) of a signal comprising at least two zones, called the decoding zone, said decoding zone comprising a plurality of samples (Pd_n) to be decoded, characterized in that the decoding device is configured to implement:

• the decoding of a group of latent value maps ( ) representative of said zone to be decoded (Zod ^j ) of said signal,
• obtaining a mask (MZo ^j ) of said zone to be decoded in said signal (I(P _n )),
• the decoding of a set of parameters (Wd _k ) representative of a neural network (NNSYN'), called a synthetic neural network comprising at least one synthetic neural layer,
• the processing of said group of decoded latent maps ( ) by said synthetic neural network (NNSYN') to produce at the output of said synthetic neural network at least said decoding area (Zod ^j ), said processing comprising the application of said at least one synthetic neural layer to transform at least one group of input latent maps (CCE), called input maps, into at least one group of output latent maps (CCS), called output maps, said application comprising obtaining a characteristic region (R ^j ) of said decoding area in said input maps from said mask (MZo ^j ), said characteristic region (R ^j ) comprising the points associated with said decoding area (Zod ^j ) and for at least one point (Pt _n ) of said decoding area, called decoding point, the processing of an associated point (Pa _n ) of said decoding point in at least one of the input latent maps (CCE) by said neural layer as a function of its proximity to a boundary of said characteristic region.

Computer program comprising instructions for carrying out the steps of an encoding process according to claim 1 or a decoding process according to any one of claims 2 to 11 when said program is executed by a computer.