Method and device for encoding and decoding images

Abstract

The invention relates to a method for encoding a signal comprising at least two zones and comprising the following steps for a zone to be encoded of said signal comprising a plurality of samples to be encoded: the construction (E22) and the segmentation (E24) of a group of latent value maps representing the signal into at least two characteristic zones, for at least one current sample of the signal associated with a position in the signal to be encoded and with a point of a characteristic zone, called the associated point: the construction (E26, E27) of a characteristic vector from the latent value maps, the application (E28) of said characteristic vector to the input of an artificial neural network, defined by a set of parameters (Wk), to obtain, at the output of said synthetic neural network, a vector (P'n) representing a decoded value of the current sample (Pn),the encoding of the latent value map group comprising the encoding of a current value of said associated point, said encoding of the current value comprising the evaluation (F221) of a proximity criterion of said associated point with respect to a boundary of said characteristic area, the construction (F221) of a neighborhood vector of said associated point from a neighborhood of said associated point in said latent value maps of said group and the evaluation of said proximity criterion, and an entropic encoding (F223, F224) of said current value as a function of said neighborhood vector, and the updating (E23, E26, E27) of at least one value of one of said latent value maps of said group and / or of at least one parameter of said synthetic neural network, as a function of an encoding performance measure, the encoding (E29) of said set of parameters of said synthetic neural network. Figure for the abstract: Fig. 5,

Description

Title of the invention: Method and device for encoding and decoding images. Prior art.

[0001] 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.

[0002] Digital videos are generally subject to source coding aimed at compressing them in order to limit the resources required for their transmission and / or storage. Numerous coding standards exist, such as the ITU / MPEG standards (H.264 / AVC, H.265 / HEVC, H.266 / VVC, etc.) and their extensions (MVC, SVC, 3D-HEVC, etc.). In these approaches, image encoding is generally performed by predicting pixels using previously encoded and then 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."

[0003] In addition to these classic approaches, approaches based on artificial intelligence, and in particular neural networks, tend to develop.

[0004] Some neural approaches, starting from an input signal, for example an image, train a so-called synthetic neural network on characteristic vectors associated with a position of a sample of the input signal to be encoded. These characteristic vectors are constructed from feature maps that may be at the resolution of the input signal, or at a lower resolution. During training, or construction, the parameters of the neural network and the values ​​of the feature maps are updated according to a performance measure, for example, a data rate-distortion type. When training is complete, i.e., when the performance measure obtained is satisfactory, the actual encoding of the synthetic neural network parameters and the values ​​of the feature maps can be performed and stored or transmitted to the decoder.The decoding of the coded signal is then carried out by applying the synthesis neural network to the feature maps.

[0005] One drawback of the classical and neural approaches described above 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 of parallelizing the decoding of this coded signal.

[0006] There is therefore a need for a solution to improve upon the classical and neural approaches described above. Summary of the invention

[0007] The invention relates to a method for encoding a signal comprising at least two zones, comprising the following steps for a zone to be encoded of said signal comprising a plurality of samples to be encoded: - the construction of a group of latent value maps representative of the signal, - obtaining a segmentation of said group of latent value maps into at least two characteristic areas, one of the at least two characteristic areas, called the associated characteristic area, being associated with said area to be coded, - for at least one sample, called the current sample, of the area to be coded, associated with a position in the signal to be coded and with a point in the associated characteristic area, called the associated point: • the construction of a characteristic vector from said latent value maps of said group, as a function of said position of said current sample, • the application of said characteristic vector to the input of an artificial neural network, called a synthetic neural network, defined by a set of parameters, to obtain, at the output of said synthetic neural network, a vector representing a decoded value of the current sample, • the encoding of a value, called the current value of said associated point, said encoding of the current value comprising the evaluation of a proximity criterion of said associated point with respect to a boundary of said characteristic area, the construction of a neighborhood vector of said associated point from a neighborhood of said associated point in said latent value maps of said group and the evaluation of said proximity criterion, and an entropic encoding of said current value as a function of said neighborhood vector, and • the updating of at least one value of one of said latent value maps of said group and / or of at least one parameter of said synthetic neural network, based on a coding performance measure, - the encoding of a group of latent value maps representative of said area to be encoded of said signal, and - the coding of said set of parameters of said synthetic neural network.

[0008] 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, said decoding comprising • the determination in said latent value maps of said group of a characteristic zone corresponding to said zone to be decoded of said signal, • for at least one value, called the current value, of a point in said characteristic zone, the evaluation of a proximity criterion for said point with respect to a boundary of said characteristic zone, the construction of a neighborhood vector for said point from a neighborhood of said point in said latent value maps of said group and the evaluation of said proximity criterion, and an entropic decoding of said current value as a function of said neighborhood vector, - the decoding of a set of parameters representative of a neural network, called a synthetic neural network, and - the processing of said group of latent value maps by the synthesis neural network.

[0009] For the purposes of this invention, encoding, or "coding," means the operation of representing a set of samples in a compact form, for example, using a digital binary stream. Decoding means the operation of processing a digital binary stream to produce decoded samples.

[0010] By "sample" of the signal, we mean 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, for example (R, G, B) or (Y, U, V). The position of the sample is located by its abscissa (x) and ordinate (y) coordinates in the image.

[0011] By "signal comprising a plurality of samples" is meant a signal in one (audio, sound), two (image) or more than two (stereoscopic image, multiscopic image, image associated with a depth map, video, etc.) dimensions. Depending on this In terms of dimensionality, the sample has one, two, or more coordinates in the signal. In the case of an image signal, the position of the sample is located by its abscissa (x) and ordinate (y) coordinates.

[0012] By "feature maps" or equivalently by "latent value maps" is meant 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.

[0013] By "characteristic data vector constructed from feature maps as a function of a position" is meant a vector consisting of one or more elements, or data, preferably discrete, the data being constructed from the feature maps at a position determined by that of the sample being processed in the signal. This characteristic vector is the one that is 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 (respectively,to decode) or in a neighborhood of it. 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.

[0014] 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 compute 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).

[0015] 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.)

[0016] By "processing by a synthetic neural network" is meant 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 (resp. decoded). This output vector may contain one or more data points representative of the sample.

[0017] By "performance measurement," we mean a measurement between at least one value of a sample to be encoded and a decoded value of said sample. The measurement may, for example, assess distortion or perceptual error. It may be performed on one or more samples (for example, a current sample, or the current image, etc.). The measurement 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 of the first group. The measurement may be a joint measurement of throughput and distortion through their weighting. As is well known in the prior art, the value of this measurement is generally minimized until a target value is reached.

[0018] Generally speaking, 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 may be carried out in a different order than that indicated in the independent encoding or decoding claim, or even in parallel.

[0019] 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. Thus, it is possible to obtain a coded representation of a region of the original signal that can subsequently be decoded independently of any other part of that signal.

[0020] 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.

[0021] According to a first characteristic, the neighborhood of said point is independent of said evaluation of the proximity criterion.

[0022] According to another feature, the neighborhood of said point is selected based on said evaluation of the proximity criterion

[0023] According to another feature, during the construction of the neighborhood vector, a component of the neighborhood vector 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 zone and to a replacement value otherwise.

[0024] According to another feature, the replacement value is dependent on the points of said characteristic zone or of the points associated with said neighborhood belonging to said characteristic zone.

[0025] According to another feature, the replacement value does not depend on the points of said feature area.

[0026] According to another feature, the neighborhood of said point is selected based on said point.

[0027] According to another feature, the construction of the neighborhood vector includes: - the identification of the points of said neighborhood belonging to said characteristic zone, called identified points, - initializing the number of components of said neighborhood vector to the number of identified points, and - the assignment of the value of each identified point to a different component of said neighborhood vector.

[0028] According to another feature, the proximity criterion of said point with respect to a boundary of said characteristic zone is a distance.

[0029] According to another feature, the decoding process further includes a step of obtaining a segmentation of said signal into at least two zones and said characteristic zone is determined from said segmentation of said signal.

[0030] According to another feature, the decoding process further includes a step of obtaining a mask of said area to be decoded in said signal and said characteristic area is determined from said mask.

[0031] According to another feature, the neighborhood vector is used to determine a context for the entropic decoding of said current value.

[0032] According to another feature, the entropic decoding of said current value includes a step of predicting said current value as a function of said neighborhood vector.

[0033] According to another feature, decoding said current value of one of said latent value maps comprises: - the decoding of a set of parameters representative of a neural network, called a predictive neural network, and - the processing of said neighborhood vector by the prediction neural network, to provide a prediction of said current value.

[0034] Correspondingly, the invention also relates to a signal coding device comprising at least two zones, characterized in that said coding device is configured to implement, for a zone to be coded, said signal comprising a plurality of samples to be coded: - the construction of a group of latent value maps representative of the signal, - obtaining a segmentation of said group of latent value maps into at least two characteristic areas, one of the at least two characteristic areas, called the associated characteristic area, being associated with said area to be coded, - for at least one sample, called the current sample, of the area to be coded, associated with a position in the signal to be coded and with a point in the associated characteristic area, called the associated point: • the construction of a characteristic vector from said latent value maps of said group, as a function of said position of said current sample, • the application of said characteristic vector to the input of an artificial neural network, called a synthetic neural network, defined by a set of parameters, to obtain, at the output of said synthetic neural network, a vector representing a decoded value of the current sample, • the encoding of a value, called the current value of said associated point, said encoding of the current value comprising the evaluation of a proximity criterion of said associated point with respect to a boundary of said characteristic area, the construction of a neighborhood vector of said associated point from a neighborhood of said associated point in said latent value maps of said group and the evaluation of said proximity criterion, and an entropic encoding of said current value as a function of said neighborhood vector, and • the updating of at least one value of one of said latent value maps of said group and / or of at least one parameter of said synthetic neural network, based on a coding performance measure, - the encoding of a group of latent value maps representative of said area to be encoded of said signal, and - the coding of said set of parameters of said synthetic neural network.

[0035] 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 encoding device is configured to implement: - the decoding of a group of latent value maps representative of said area to be decoded of said signal, said decoding comprising • the determination in said latent value maps of said group of a characteristic zone corresponding to said zone to be decoded of said signal, • for at least one value, called the current value, of a point in said characteristic zone, the evaluation of a proximity criterion of said point relative to a boundary of said characteristic zone, the construction of a neighborhood vector of said point from a neighborhood of said point in said latent value maps of said group and the evaluation of said proximity criterion, and an entropic decoding of said current value as a function of said neighborhood vector, - the decoding of a set of parameters representative of a neural network, called a synthetic neural network, and - the processing of said group of latent value maps by the synthesis neural network.

[0036] 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.

[0037] 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 can use any programming language and be in the form of source code, object code, or code intermediate between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0038] The invention also relates to a computer-readable information or recording medium comprising the computer program instructions mentioned above. The information or recording medium may be any entity or device capable of storing programs. For example, the medium may include a storage means, such as a ROM, for example a CD-ROM or a microelectronic circuit ROM, or a magnetic recording means, for example a floppy disk or a 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 may be transmitted via an electrical or optical cable, by radio link, by wireless optical link, or by other means.

[0039] The program according to the invention can in particular be downloaded onto an Internet-type network.

[0040] Alternatively, each information or recording medium may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of a process according to the invention. Brief description of the figures

[0041] 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: - [Fig. 1] schematically represents a coding device according to a first embodiment of the invention, - Figure [Fig. 2] schematically represents a decoding device according to the first embodiment of the invention, - Figure 3 illustrates an example of a synthetic artificial neural network used within the scope of the invention, - Figure 4 illustrates an example of an artificial predictive neural network used in the context of the invention. - [Fig. 5] is a flowchart representing an example of a coding process which can be implemented by the coding device of [Fig.1], - Figure 6 illustrates a coding method used in an embodiment of the invention, - [Fig.7] is a flowchart representing an example of a decoding process that can be implemented by the decoding device of [Fig.2], - Figure 8 illustrates a decoding process used in one embodiment of the invention, - [Fig.9] is a flowchart representing a method for encoding feature cards that can be implemented by the encoding device of [Fig.1] and by the encoding process of [Fig.5], - [Fig. 10] is a logic diagram representing a method for decoding feature maps which can be implemented by the decoding device of [Fig.2] and by the decoding process of [Fig.7]. - Fig.11 and Fig.12 illustrate the formation of a neighborhood vector used in a method of encoding or decoding feature maps. - Figure 13 illustrates a first variant of vector formation neighborhood. - Figure 14 illustrates a second variant of vector formation. neighborhood. - Figure 15 illustrates a third variant of vector formation neighborhood. - Figure 16 illustrates a fourth variant of vector formation neighborhood. Detailed description of the invention

[0042] Figure 1 schematically represents, according to a first embodiment, an ENC coding device for at least one zone of a signal (I(Pn)). In the example described here, the signal I(Pn) is segmented into a set of J zones, and all zones are coded independently. Alternatively, only one zone or only a few zones may be coded independently.

[0043] 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 transformation module, an XTR data extraction module, a TT processing and quantification module, an NNSYN module corresponding to a synthetic artificial neural network, an NNC neural network coding module capable of encoding the synthetic neural network and optionally the prediction neural network, an FMC feature map coding module, an EVAL module for evaluating coding performance, and an MAJ update module.

[0044] 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.

[0045] The ENC coding device of [Fig. 1] receives as input a signal consisting of a succession of samples to be encoded, denoted Pn, for example, a temporal sequence of sound samples, or a set of image data denoted I(Pn). In this second case, the image signal I(Pn) can represent a two-dimensional image, or a plurality of two-dimensional images (video, color components, stereoscopic components, multiscopic components, etc.). Pn 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 (R), green (G), blue (B) component, or, alternatively, a luminance (Y) component and at least one chrominance (U, V) component. The location of each pixel is defined by its x and 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.

[0046] The SEG module performs a segmentation Sg of the sequence of samples to be encoded Pn into J (greater than or equal to 2) zones Zo'. This segmentation operation allows the samples to be grouped into J different homogeneous sets according to one or more predefined criteria. For example, in the case of a temporal sequence of sound samples, the segmentation operation makes it possible to obtain sequences of units corresponding to silences, noises, phonemes, words, etc. Similarly, in the case of an image signal I(Pn), the segmentation operation makes it possible to group the pixels Pn of the image signal into J homogeneous zones according to criteria, notably intensity or spatial criteria. For example, the segmentation operation can make it possible to identify in the image signal I(Pn) two different zones (J=2), one corresponding to the background of the image and the other to the foreground of the image.Other examples of segmentation will be presented later with reference to the figure. Segmentation Sg, for example, is represented as a set of J masks MZo', each allowing the identification of a zone Zo*, each mask being associated with a value different from that associated with the other masks.

[0047] 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' masks (for example, using the JPEG-LS algorithm defined by the international standard ISO / IEC 14495-1) or, alternatively, by independently encoding each MZo* mask, or, in yet another alternative, by independently encoding the contours of the J MZo* masks. The encoded segmentation is denoted Sgc.

[0048] The NNC module performs a simulation of coding, followed by decoding, of the parameters of the NNSYN neural network for the evaluation module.

[0049] The GEN feature map generation module is configured to generate a plurality of M feature maps, denoted FM;.

[0050] In one embodiment, the SE module performs a transformation of the first group of FM characteristic maps; to generate a second group of FMS characteristic maps; at the same resolution as the input signal.

[0051] The optional SE module can perform quantization of the data extracted from this set of M FM 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 may consist of integer values, and the quantization system performs a simple rounding of a real value to an integer. In another example, quantization consists of multiplication by a given value followed by rounding. Then the SE module performs a transformation of the values ​​of at least one of the feature maps, for example, oversampling, interpolation, or a 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, in this method, the feature maps that are encoded can have a lower resolution than the images to be encoded, while the maps from the second group, which are used to construct the feature vectors, have the same resolution as the image sequence, thus facilitating value extraction.

[0052] In one embodiment, the SE module is absent; in this case, the values ​​which will be used to construct the characteristic vector are extracted from the first group of characteristic maps.

[0053] The XTR module performs value extraction from the FMSi (or FM; according to one of the embodiments described above) feature maps for a current sample Pn to be encoded, based on its coordinates in the input signal. For example, if one seeks to encode the sample Pn at the coordinates (xn, yn) of an input image, the XTR module performs value extraction from the maps at positions imposed by the coordinates (xn, yn).

[0054] In one embodiment, the extracted values ​​constitute the vector Zn. Zn is an L-tuple, that is, it contains L elements, or data z;. For example, in one embodiment, L=M, meaning, for example, that only one value is extracted for each feature map FM;. In another embodiment, L>M, meaning, for example, that several values ​​are extracted for at least one feature map FM;. The dimension L of the vector depends on the topology of the NNSYN synthesis neural network and, more particularly, on the topology of the input layer of this NNSYN synthesis neural network.

[0055] The vector Zn of index n refers to the characteristic vector of the pixel P'n.

[0056] In one embodiment, the optional TT module processes the extracted values ​​to generate the vector Zn. The TT module can quantify the data extracted from the set of feature maps. The processing may include other operations, such as filtering, scaling, etc. In particular, if the SE module is not used and if 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.

[0057] It should be noted that at least one of the SE or TT modules must perform a quantification of the characteristic maps.

[0058] The NNSYN module is a synthetic neural network defined by K parameters Wk, capable of processing the input vector Zn, or L-tuple, to generate as output a second vector representative of the sample Pn to be encoded. The synthetic neural network is In one embodiment, an MLP (Multi-Layer Perceptron) consists of an input layer adapted to the input format (the L-tuple), optionally one or more hidden layers, and an output layer adapted to the output format of the output vector, generally a vector containing A elements. In one embodiment, A is equal to 3 and the output vector is the triplet (R, G, B) of the encoded and then decoded pixel P'n.

[0059] The NNC module performs the coding of the synthesis neural network, in particular its parameters Wk. Optionally, the NNC module also performs the coding of the prediction neural network, ARM, in particular its parameters Ob. During the coding training or construction process—that is, as long as the performance evaluation step is not satisfactory—the NNC module performs a coding simulation, followed by decoding, for the evaluation module. Subsequently, it performs the actual coding of the parameters of the synthesis neural network Wk and optionally the coding of the prediction neural network, ARM. The coded parameters are denoted Wck and Ocb. In a known manner, the coding simulation can be identical to the actual coding, or it can perform an approximation of it.

[0060] The FMC module performs the encoding of the FM maps; that is, the values ​​of the feature maps of the first group (excluding the maps of the second group, which optionally result from oversampling by the SE module). During the encoding training, or construction, process—that is, as long as the performance evaluation step is not satisfactory—the FMC module performs a coding simulation, followed by decoding, for the evaluation module. Subsequently, it performs the actual encoding of the FM map values.

[0061] The FMC module determines, taking into account the Sg segmentation performed by the SEG module, an SgL segmentation of each of the FM characteristic maps; in J zones ZolJ- Thus, the FMC module uses the Sg segmentation defined in the domain of the signal to be encoded to obtain a segmentation in the latent domain of the FM characteristic maps; Obtaining this segmentation in the latent domain depends on the transformation transform used to go from the domain of the signal to be encoded to the latent domain. Thus, the SgL segmentation in the latent domain results from a calculation that simultaneously takes into account the location of the Sg segmentation zones in the signal to be encoded and the transformation transform. For example, if the latent domain and the signal to be encoded have the same resolution, then the segmentation in the latent domain is identical to the segmentation in the domain of the signal to be encoded.According to another example, if a latent has a resolution lower than that of the signal to be encoded, then the boundary between two areas of the segmentation in the domain of this latent lies between the points in the latent domain whose samples are colocalized, i.e. The corresponding values ​​in the signal to be encoded belong to two different regions of the SgL segmentation. Each FM map is represented as a set of J zones 2oL', each of the corresponding 2oL' zones in the FM map to region j of the SgL segmentation. The encoding of the FM map values ​​is carried out taking the SgL segmentation into account. More precisely, each of the J 2oL' zones is encoded successively, for example, in the order of their indexing. The encoding of a 2oL' zone comprises the successive encoding, for each FM map, of the point values ​​of the 2oL' zone within the FM map. Thus, each of the 2oL' zones is encoded independently of the other zones. The encoded data of each encoded zone 2oL' in each of the FM feature maps are denoted by J. Each 2oL' zone undergoes entropic encoding. Thus, the encoding of the M FM maps corresponds to the independent encoding of the J zones of the M FM maps.

[0062] In a known manner, 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 coding that takes into account the neighborhood of a value to be coded in the feature map. As will be detailed later, the FMC module can include a predictive neural network (PNN).

[0063] 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.

[0064] The MAJ module performs an update of the values ​​of the FM cards; and / or the parameters of the neural network(s) to be encoded, according to the results of the performance function.

[0065] Fig. 2 schematically represents a decoding device DEC of a decoding zone Zodj of a signal, called the decoding zone, said decoding zone Zodj comprising a plurality of samples Pdn to be decoded.

[0066] This DEC decoding device includes an NND neural network(s) decoding module capable of decoding the NNSYN' synthesis neural network and optionally the ARM' prediction neural network, a SEGD segmentation decoding module, an FMD feature map decoding module, an XTR' data extraction module, an SE' inverse transformation module, and a TT' inverse processing and quantization module.

[0067] The DEC decoding device produces at output a decoded image comprising at least the decoded area, denoted Zodj (Pdn), comprising a plurality of decoded samples Pdn.

[0068] The DEC decoding device of Figure 2 receives as input the encoded segmentation Sgc and a group of encoded data pMc^

[0069] The DEC decoding device of [Fig.2] also receives as input the Wck encoded parameters of the NNSYN' synthesis neural network and optionally the Ocb encoded parameters of the ARM' synthesis neural network.

[0070] The parameters of the NNSYN' synthesis neural network decoded by the NND module are denoted Wdk. The parameters of the ARM' prediction neural network decoded by the NND module are denoted Odb.

[0071] The SEGD module decodes the MZo' mask from the Sgc encoded data.

[0072] According to embodiments as described for the encoder: - The segmentation map corresponding to the J MZo* masks is decoded and the MZo* mask is extracted from the segmentation map. - The MZo* mask is decoded directly when it has been previously encoded in the Sgc encoded data independently of other masks.

[0073] The FMD module constructs M decoded maps using the MZo* mask and the encoded data pM(A From the mask MZo* in the signal domain to be decoded, the FMD module determines the corresponding ZoLj zone in the latent domain, then the FMD module decodes the pjjcJ data to obtain the values ​​of the points in this ZoLj zone in each of the pfyfd-- maps

[0074] In one embodiment, the SE' module performs a transformation of the first group of decoded feature maps pMd- to generate a second group of feature maps at the same resolution as the signal to be decoded, denoted pjvfS'i-. 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 has simply rounded the real values ​​submitted to it. 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 the quantizer Q. Then, the SE' module performs a transformation of the feature map values, including, for example, oversampling, interpolation, filtering, etc., similar to that performed by the encoder.Following the transformation, a map of . The transformed characteristics of the second group are of the same resolution as the images of the sequence to be decoded.

[0075] 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.

[0076] The XTR' module is identical to the XTR module of Figure 1. It performs an extraction of values ​​from the M characteristic maps (or pMS'] according to one of the embodiments described above), for a sample Pdn to be decoded, as a function of its coordinates in the signal to be decoded.

[0077] In one embodiment, the extracted values ​​constitute the vector Zdn. Zdn is an L-tuple, that is to say, it comprises L elements, or zd data;

[0078] In one embodiment, the optional TT' module processes the extracted values ​​to generate the Zdn vector. 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.

[0079] The NNSYN' module is a so-called synthetic neural network, defined by K parameters Wdk, capable of processing the input vector Zdn, or L-tuple, to generate as output a second vector representing the sample Pdn 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 Pdn. The NNSYN' module has the same structure as the NNSYN module, and its parameters are either identical if the encoding of its parameters Wk is lossy, or different if the encoding is lossy.

[0080] When all the Pdn samples of the signal have been decoded, we have a reconstructed signal Zodj(Pdn) of the ZoJ area.

[0081] Furthermore, the DEC decoding device can successively be implemented to decode all the Zodj areas in order to reconstruct all the samples to be decoded of the signal, i.e. of the image I(Pdn).

[0082] The DEC decoding 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 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 concerned, in particular as described below, when these instructions are executed by the processor. The DEC device can also comprise a plurality of processors, the processors being dedicated to the parallel decoding of image areas.

[0083] Figure 3 illustrates an example of a synthetic artificial neural network used for encoding and decoding in one embodiment of the invention.

[0084] 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, comprising 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.

[0085] A vector representation of a current sample (a vector Zn or Zdn from the FM / FMSi feature maps in the encoder or FJvJd-j / FMS'- in the decoder) is applied as input (i.e., to an input layer) to the NNSYN or NNSYN' synthetic artificial neural network. The synthetic artificial neural network outputs a vector, according to one embodiment, representing the color components (R, G, B) constituting a color pixel of an image or an area of ​​an image.

[0086] The concatenation of all these reconstructed pixels in an image constitutes the image or the area of ​​the image decoded, or reconstructed.

[0087] At the encoder, the NNSYN synthetic artificial neural network is trained on the entire image, so as to minimize the differences between the input representation of the current image I(Pn) 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.

[0088] 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.

[0089] Figure 4 illustrates an example of an artificial predictive neural network used for encoding (ARM) and decoding (ARM') for feature maps, within the framework of embodiments of the invention.

[0090] The artificial neural network for prediction used in encoding, ARM, and the artificial neural network for prediction used in decoding, ARM', are defined by an identical structure, comprising 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.

[0091] A vector representation of a current neighborhood (a vector Cn or Cdn from the FM feature maps; at the encoder or FMd1 at the decoder) is given as input (i.e. on an input layer) of the ARM (at encoding) or ARM' (at decoding) artificial prediction neural network.

[0092] The artificial predictive neural network behaves as a function which produces as output a prediction of the current value of the feature map being processed, which can be in the form of a prediction value or probability data.

[0093] According to one embodiment, at the encoder, the network implements a function fv that provides an expected mean and / or variance (p,o) for the current value Vn of the FM card to be encoded. These statistical values ​​are used to perform the entropic coding of this value. For example, if the function produces a mean, this mean is subtracted from the current value, and only the difference is entropically coded, this mean being considered a prediction of the current value. Alternatively, if the function produces a mean and a variance, the mean is subtracted from the current value, and the difference is coded using an entropic coding adapted to the variance, for example, by quantifying said variance into a predetermined set of variances and associating a type of entropic coding with each quantified variance value.At the decoder, the network implements a function fv that provides an expected mean and / or variance for the current value Vdn of the card to be decoded. These statistical values ​​are used to perform entropic decoding of this value. For example, if the function produces a mean, the current value is decoded by the entropic decoder, and this mean is added to the current value. Alternatively, if the function produces a mean and a variance, the current value is decoded by the decoder using an entropic decoding method adapted to the variance, for example, by quantizing the variance into a predetermined set of variances and associating a type of entropic decoding with each quantized variance value.

[0094] According to another embodiment, the neural network can produce the expected probability (pr) of each possible value in the current sample. In this case, entropic encoding or decoding will be adapted to this probability (as is known for Huffman or arithmetic entropic coding).

[0095] At the encoder, the ARM artificial neural network for prediction is trained on the entire image so as to minimize the size (in number of bytes or bits) of the encoded data. The EVAL module performs a performance measurement in this regard. It should be recalled that the overall performance measurement focuses on minimizing the distortion between encoded and then decoded images I(P'n) and the input images I(Pn), while minimizing the encoding rate. In one embodiment, the feature maps are losslessly encoded using entropy coding. In this case, the encoding of the feature maps affects the rate, but not the distortion of the encoded images. In another embodiment, if the feature maps are lossy encoded, the encoding of the feature maps affects both the throughput and the distortion.

[0096] Once training is complete, the B parameters Ocb of the network are encoded, either losslessly, in which case the neural network ARM' is identical to ARM, or with losses, in which case the network ARM' may be slightly different from ARM.

[0097] Fig. 5 is a logic diagram representing an example of a coding process that can be implemented by the coding device of Fig. 1 when the NNSYN synthesis neural network is that shown in Fig. 3.

[0098] According to this embodiment, the signal is a two-dimensional image, each sample to be coded is therefore a pixel Pn with coordinates (xn, yn).

[0099] The encoding takes place in three main phases:

[0100] In a first phase, called the segmentation phase, the segmentation Sg of the input signal I(Pn) into J (greater than or equal to 2) zones Zo' is carried out.

[0101] According to 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 whose boundaries are horizontal and vertical. Such a division corresponds to the concept of "tiles" implemented by coding standards such as H.264 / AVC, H.265 / HEVC and H.266 / VVC.

[0102] According to a second example of segmentation, the image is divided into J zones corresponding to samples (or blocks) traversed 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.

[0103] According to a third example of segmentation, the image is divided into semantic zones, for example, 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 should be noted that there are no restrictions related to the type of segmentation algorithm used.

[0104] Advantageously, such a division into zones allows, during the subsequent decoding of the signal I(Pdn), to parallelize the decoding of the zones to be decoded Zodj by distributing the decoding load equally on each of the decoders and / or processors.

[0105] In a second phase, called the construction phase, learning is performed to determine, for an input signal I(Pn), the values ​​of the FM maps; and the parameters Wk and optionally Ob to optimize an overall cost function. The learning is, for example, performed by gradient descent, followed by an update of the parameters of the NNSYN synthesis neural network, the values ​​of the FM feature maps; and optionally the ARM prediction neural network. As is known in the state of the art, the cost function can be rate-distortion, rate-distortion, or perceptual. To measure the throughput R, it is necessary to simulate the encoding of the J zones ZoLj of the FMb maps and then measure the associated encoding throughput (the size of the B2 stream). In one embodiment, the encoding of parameters Wk and / or Ob is not simulated because their influence is less significant than that of the feature maps. In another embodiment, the encoding of parameters Wk and / or Ob is also simulated, and the associated throughput (the size of the Bl stream) is measured.To measure the distortion D, it is necessary to simulate the encoding and then the decoding of at least a part of the image I, to obtain at least one pixel P'n resulting from a simulation of encoding and then decoding, and then to measure the difference between this part of the input image I(Pn) and a corresponding part of the encoded and then decoded image I(P'n).

[0106] Then, during a third phase, called the coding phase, the segmentation Sg, the data of each zone ZoL^ and its parameters Wk and possibly Ob are encoded to produce the coded values ​​Sgc and Wck (and possibly Ocb) before transmission or storage. These constitute the compressed representation of the input signal I(Pn), this compressed signal being able to be decoded zone by zone.

[0107] We will now describe the steps of a coding process according to one embodiment of the invention.

[0108] During a step E20, a signal I(Pn) to be coded, comprising a plurality of N samples Pn, is provided as input to the process.

[0109] During a step E21, the segmentation Sg of the input signal I(Pn) into J (greater than or equal to 2) zones Zo' is carried out.

[0110] During a step E22, the M FM maps; of the first group. Subsequently, the parameters Wk of the synthesis neural network NNSYN and the values ​​of the FM maps i, and possibly the parameters Ob of the prediction neural network, must be optimized during the construction phase.

[0111] According to one embodiment, the FM cards; are of the same resolution as the input signal I(Pn) and therefore each have the same number of values ​​N as there are samples Pn to be coded.

[0112] According to one embodiment, the FM cards; have a resolution less than or equal to that of the input signal I(Pn) 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 card; has the resolution of the images and each subsequent card has half the resolution of the previous one.

[0113] According to one embodiment, several FM cards; are of the same resolution, lower than that of the input signal I(Pn).

[0114] According to one embodiment, the FM maps are transformed to provide a second group of transformed feature maps FMS. 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. The maps of the second group are not coded; they are used solely for constructing the feature vectors.

[0115] According to one embodiment, the FM cards are initialized with predefined constant values.

[0116] According to another embodiment, the feature cards are initialized by a set of random real values.

[0117] The FM feature cards; of the first group are subsequently updated, or refined, during an E23 step, by the encoder update module during its learning.

[0118] During step E24, the J zones of the FM maps of the first group are determined and 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 may be identical to the actual encoding, but it may also be different (for example, simplified). For this encoding, a technique for predicting a feature map value based on its neighborhood is used, as will be described, for example, in support of [Fig. 9]. In one embodiment, the structure and Ob parameters of the prediction 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.

[0119] In one embodiment, the J zones ZoLj of the FM maps are encoded in the indexing order of the J zones and in a predefined order of the associated feature maps (i.e., FM1, FM2, FM3, FM4, FM1, ...) and the values ​​of each map, for example, lexicographical. Each zone ZoLj of the FM maps undergoes entropic coding.

[0120] The entropic coding of the set of zones ^oL-* the set of FM cards; produces a compressed stream B2 whose throughput is measured subsequently during a step E29.

[0121] During a step E25, according to one embodiment, the M cards of the first group FM; are transformed by the SE module to generate cards of the second group FMS; at the resolution of the images of the input sequence.

[0122] According to one embodiment, M FMS cards are generated.

[0123] According to one embodiment, each FM card; is transformed into an FMS card;.

[0124] According to one embodiment, at least one FM card; is of lower resolution to that of the images in the sequence to be coded, and the transformation operation includes Oversampling is performed so that the transformed FMS map contains the same number of samples as the images in the sequence. Oversampling involves adding values ​​to the FMS maps to achieve the resolution of the input sequence images. It can be simple (nearest neighbor replication) or involve interpolation (linear, polynomial, filtered, etc.).

[0125] During step E26, values ​​are extracted by the XTR module from the FM cards; or possibly FMS cards; transformed. This extraction is performed according to the coordinates (xn, yn) of the sample Pn of the input signal. It can also be performed according to the resolution of the card considered.

[0126] According to one embodiment, the characteristic Zn vector results directly from this extraction.

[0127] The samples to be coded are, for example, processed sequentially, from n=1 to n=N.

[0128] According to one embodiment, during a step E27, the characteristic vector Zn is constructed by the TT module from the values ​​extracted from the FM or FMS maps for each sample Pn with coordinates (xn, yn) of the input signal. The processing may include quantization of the values ​​extracted from the FM maps or of the resulting vector Zn, if necessary. The processing may include other operations, such as filtering, scaling, the application of any function, preferably monotonic, etc.

[0129] In one embodiment, Zn is an M-tuple (zb z2,..., zM), consisting of the values ​​of the FM; or FMS; maps located at the coordinates (xn, yn) of the current pixel Pn, as will be illustrated in support of [Fig.6].

[0130] In one embodiment, Zn is an M-tuple constructed from values ​​taken from the FMi maps at coordinates that may be different for each map. For example, if the FMi maps are at different resolutions because they have been undersampled, the coordinates are adjusted (by scaling) to match the resolution of each map.

[0131] During a step E28, the vector Zn is processed by the NNSYN synthesis neural network to generate as output a vector representative of the sample Pn to be coded, according to an embodiment the triplet (R, G, B) of the sample P'n (the sample Pn coded then decoded).

[0132] The structure and parameters Wk of the synthesis neural network are initialized, for example, during the first iteration of this step. These parameters are by the updated, or refined, during the construction phase, during subsequent iterations of the process.

[0133] 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).

[0134] 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.

[0135] During step E29, the parameters Wk of the synthesis neural network NNSYN and the parameters Ob of the prediction neural network ARM, if it exists, are quantized and encoded. During the construction phase, this operation is a coding simulation. During the encoding phase, this operation is an actual encoding, and the encoded values ​​constitute the Bl stream. The simulation may be identical to the actual encoding, but it may also be different (for example, simplified). Any known technique may be used for this purpose, for example, the neural network coding standard proposed by the MPEG-7 Part 17 standard, also called NNR (Neural Network Representation). Note that in this case, the amount of degradation that the encoding introduces to the weights Wket, optionally Ob, must be chosen.

[0136] During an E30 step, a performance measure is evaluated.

[0137] To this end, the coding simulation rates associated with the feature maps of the first group (simulation of the B2 flow by coding the J zones jTjyji of the FM maps;) and optionally with the parameters of the neural network(s) (simulation of the B1 flow by coding the parameters Wket optionally Ob) are measured.

[0138] In one embodiment, the cost function is of the rate-distortion type, 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 (for Structural SIMilarity) or MSSSIM (for Multi-scale Structural SIMilarity). In one embodiment, R is the simulated rate of the B1 stream; in another embodiment, R is the total rate used to encode this image, i.e., the sum of the simulated rates of B1 and B2. L is a parameter that controls the rate-distortion trade-off. Other cost functions are possible.

[0139] As long as the cost function has not reached its minimum, or a maximum number of iterations of the cost function minimization algorithm has not been reached, the performance measurement is not satisfactory, and the process is repeated from step E23. This minimization can be performed by a known mechanism like a gradient descent with parameter updates during step E23 for feature map values ​​and E26 / E27 for network(s) parameters.

[0140] During a EF 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 Wk synthesis neural network parameters and feature maps is available, the B1 and B2 streams can be constructed from it. According to another embodiment, the actual encoding of the updated parameters of the synthesis neural network Wk and the values ​​of the feature maps FM, and optionally of the prediction neural network Ob, is performed at this stage to produce the encoded parameters Wck (optionally Ob) and pMcj, which constitute streams B1 and B2. Furthermore, at this stage, the encoding of the segmentation Sg is performed to produce an encoded segmentation of the J masks MZoc', which constitutes stream B3.

[0141] The B2 stream comprises sub-streams corresponding to each area of ​​the feature maps. During the coding of each latent area, it is possible, in a preferred embodiment, to place information indicating the location of the sub-streams within the overall stream at an identifiable point in the stream. This location information can be: - a flow pointer (which indicates the starting address of each sub-flow in the overall flow), or - a marker (a series of bits otherwise prohibited, which allows traversing the stream to find the beginning of each sub-stream), or - any other means of identifying a sub-part of a coded stream.

[0142] Streams B1 and B2 can be multiplexed and / or concatenated to produce a final stream. In one embodiment, stream B3 of the coded segmentation and stream B1 of the coded parameters of the neural network(s) are stored or transmitted before stream B2, so that they can be decoded before stream B2.

[0143] Figure 6 illustrates a coding method used in one embodiment of the invention.

[0144] In this embodiment, there are 4 FM cards generated. In a preferred mode, there are 7.

[0145] The first FMi card has the same resolution as the image I(Pn), and therefore has variable WxH dimensions, where W represents the width of the image in pixels, and H its height. The second FM2 card has half the resolution (in each dimension) of the FMi card. Each additional card has half the resolution of the preceding card. This This structure allows us to reduce the number of variables in feature maps, which facilitates coding and learning while minimizing coding costs.

[0146] The FM2 map is oversampled by the SE module by a factor of 2 in each dimension, according to a method illustrated in support of [Fig. 6]. The FM3 map is oversampled by a factor of 4 in each dimension, and the FM4 map by a factor of 8 in each dimension. The FMi map is not affected by the oversampling. (FMS^FMj).

[0147] The resulting FMS maps are of the same resolution as the image I(Pn), and therefore each have WxH values, where W represents the width of the image in pixels and H its height (N=WxH).

[0148] Other types of structure are possible, for example one can use a different reduction rate of one half between the cards (one quarter, or one third, etc.).

[0149] In this embodiment, the vector Zn is a 4-tuple (zi...z4) consisting of the values ​​extracted from the FMS maps located at the coordinates (xn, yn) of the current pixel Pn. The vector Zn, consisting of the extracted (quantized) values ​​from the FMS maps, 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 (xn, yn) of the color components (R', G', B').

[0150] In another embodiment, not shown, the Znest vector is extracted directly from the FMi layers, with positions recalculated according to the size of the maps, and then the extracted values ​​are optionally processed and quantified after extraction.

[0151] [Fig.7] is a logic diagram representing an example of a decoding process for a Zodj area to be decoded which can be implemented by the DEC decoding device of [Fig.2] when the synthesis neural network NNSYN' is for example that presented in [Fig.3].

[0152] During a step F20, the Bl stream, a portion of the B2 stream (that corresponding to the feature maps of the Zodj zone), and the B3 stream are extracted from the encoded stream. They contain, respectively, Wck parameters, optionally Oc b parameters, an Sgc segmentation, and encoded representations of the M maps of the first group representative of the Zodj zone to be decoded.

[0153] During an F21 step, an MZoj segmentation mask is generated by decoding the values ​​of the encoded segmentation.

[0154] During a step F22, the M pjyyj cards are generated by decoding the FMc' values. For this decoding, the MZoj mask and a prediction technique are used. of a feature card value by its neighborhood, as will be described, for example, in support of Figure 10. In one embodiment, the pjyppisont cards are decoded in an order, for example lexicographically, (pM^, FMriy' • • FMd{P ct 'cs values ​​of each card in a predefined order, for example in rows then in columns, taking into account the MZof mask

[0155] According to embodiments as described for the encoder: - The pj^i cards have the same resolution as the Zodj(Pdn) signal to be reconstructed, that is to say they contain N=WxH values. - The p^^ maps are at a resolution less than or equal to that of the Zodj(Pdn) signal to be reconstructed. - Several FM cards have the same resolution, lower than the signal resolution.

[0156] In step F23, according to one embodiment, the M maps of the first group PM(Ji) are transformed by the SE' module to generate maps of the second group FMS' at the resolution of the input images. This step is similar to step E25, which was described for the encoder in support of [Fig. 5], and the embodiments apply. In particular:

[0157] According to one embodiment, M p^j^d cards are generated.

[0158] According to one embodiment, each card pj^j^i is transformed into a card pi^^'].

[0159] According to one embodiment, at least one pp^ppi card has a resolution lower than to that of the images of the image to be encoded and the transformation operation includes a Oversampling is performed so that the transformed map contains the same number of samples as the input image. Oversampling involves adding values ​​to the FMS'j maps to achieve the resolution of the input image. It can be simple (nearest neighbor replication) or involve interpolation (linear, polynomial, filtered, etc.).

[0160] The transformation may optionally include inverse quantization of the extracted values, if necessary. However, inverse quantization is not mandatory.

[0161] During step F24, values ​​are extracted by the XTR' module from the transformed pj^i or possibly FM$'j maps. This extraction is performed based on the coordinates (xn, yn) of a sample Pn of the input signal. It can also be performed based on the resolution of the map considered. This step is similar to step E26, which has been described for the encoder in support of [Fig. 5], and the embodiments apply. In particular:

[0162] In one embodiment, Zdn is an M-tuple (zb z2,..., zM), consisting of the values ​​of the maps or located at the coordinates (xn, yn) of a current pixel Pdn, as will be illustrated in support of [Fig.8].

[0163] The samples to be decoded are, for example, processed in sequential order, relative to the mask MZo'.

[0164] According to one embodiment, in step F35, a vector Zdn is constructed by the module TT' from the values ​​extracted from the pfyjdi maps of the first group or the maps of the second group, for each sample Pdn of coordinates (xn, yn) to be decoded, as a function of the coordinates (xn, yn). This step is similar to step E27, which was described for the encoder supporting [Fig. 5], and the described embodiments apply. The extraction may include inverse quantization of the extracted values ​​or of the constructed vector Zdn, if necessary.

[0165] During step F26, the Wdk parameters of the NNSYN' synthesis neural network and optionally the Odb parameters of the prediction neural network are generated by decoding the Wck and Ocb values ​​of the Bl 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, i.e., it has the same structure and parameters, except for the encoding, which can be performed with or without loss. Similarly, the ARM' prediction neural network, if used to decode the feature maps, is similar to the ARM prediction network, i.e., it has the same structure and parameters, except for the encoding, which can be performed with or without loss.

[0166] According to one embodiment, stream B1 is decoded before streams B2 and B3, so that the synthesis neural network NNSYN' and optionally the prediction neural network ARM' are 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.

[0167] During step F27, the vector Zdn is processed by the NNSYN' synthesis neural network to generate as output a second vector representing the sample Pdn to be decoded, according to an embodiment a triplet which is injected into the image of the decoded area Zodj(Pdn) at the positions (xn, yn) of the color components (Rd, Gd, Bd). This step is similar to step E28 which was described for the encoder in support of [Fig. 5].

[0168] When all the samples of the signal have been processed, the decoded signal corresponding to the Zodj area (Pdn) is available.

[0169] The decoding of one area of ​​the image has been described. The decoding of a second area can be carried out successively or in parallel, since the areas are coded independently of each other.

[0170] If the zones are decoded in parallel, it is necessary to have information indicating the precise position of the sub-streams (of latent data) corresponding to the different zones (for example, sub-stream pointers). On the other hand, if the zones are decoded sequentially, one can move from one sub-stream to another to decode the zones one after the other, without any specific indication.

[0171] Figure [Fig.8] illustrates a method for decoding a Zodj zone used in an embodiment of the invention.

[0172] In this embodiment, there are 4 decoded cards. In a Their preferred mode of operation is 7 in number.

[0173] In this embodiment, the decoded cards are representative 1 only from the Zodj zone. In other words, only the data from the ZoLj zone (corresponding in the latent domain to the Zodj zone) were decoded.

[0174] In this embodiment, the first map pM^J has the same resolution as image I, and therefore comprises WxH variables, where W represents the width of the image in pixels, and H its height. The second map p^d^ has half the resolution (in each dimension) of the map p^jF. Each additional map has half the resolution of the preceding map. This structure reduces the number of variables in the feature maps, thus facilitating decoding while minimizing the coding cost.

[0175] The p^-Jd- map is oversampled by a factor of 2 in each dimension, according to any oversampling method available to a person skilled in the art. The pi^pd^ map is oversampled by a factor of 4 in each dimension, and the p]\qdJ ^' map by a factor of 8 in each dimension.

[0176] The pMS'^ cards have 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.

[0177] In this embodiment, the vector Zdn is a 4-tuple (zi...z4) consisting of the values ​​of the pM$'j maps located at the coordinates (xn, yn) of the current pixel Pdn. The vector Zdn is optionally dequantized and then processed by the NNSYN' synthesis neural network to generate as output the triplet (R, G, B) representative of the sample Pdn to be decoded. The triplet (R, G, B) is inserted into the decoded image I(Pdn) at coordinates (xn, yn) in the color components (Rd, Gd, Bd).

[0178] Figure 9 is a flowchart representing a method of encoding cards characteristics which can be implemented by the coding device of [Fig.1] and by the coding process of [Fig.5].

[0179] These steps constitute sub-steps of step E24 described previously in support of [Fig.5]. They are intended to encode a current value Vn of a ZoLj zone of an FM feature map; of the first group being processed using values ​​from a neighborhood of the current value Vn.

[0180] In a substep E241, a neighborhood vector Cn is established, comprising values ​​close to the value Vn. As will be illustrated later with reference to Figures 11 to 16, these neighboring values ​​may be located in the same map and / or in a different map from the plurality M of FM maps. This neighborhood vector consists of a number C of values ​​(for example, C=4, C=8, C=9, or C=11 in the examples in Figures 11 to 16), or data, corresponding to neighborhood values ​​of the current value Vn when these neighborhood values ​​belong to the ZoLj zone, or to replacement values ​​otherwise. These values ​​must be known to the encoder and the decoder; therefore, they must be in a causal neighborhood of the value Vn.

[0181] According to a first embodiment, these values ​​are used to determine the context of an entropy encoder to encode the current value during an E244 step. This encoder can be a CAB AC (Context-adaptive binary arithmetic coding) type 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 whose compression is lossless. 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. Adaptive or non-adaptive 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 non-zero neighbor values ​​and associate a context with each number. Alternatively, one can perform comparisons between several neighbor values ​​and associate a given context with an ordering configuration among the neighbor values, for example, by sorting the neighbor values ​​in ascending order and associating a context with each possible order.

[0182] In a second embodiment, the neighborhood is used to predict, during a step E242, the current value from an autoregressive model. recalls that an autoregressive model predicts a sample from a series based on its past values. 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 E244.

[0183] In a third embodiment, as illustrated in Figures 11 to 16, a predictive neural network ARM is used during step E243 to predict the statistical characteristics of the variable to be coded. The neighborhood vector is given as input to the ARM network, which provides an output prediction of the current value. According to one embodiment, the ARM network behaves like a function fv as described in Figure 4, which provides a set of statistical parameters (mean, variance, median, etc.) used to entropically encode the current value. The role of this ARM module is to predict the current value as accurately as possible for all the values ​​Vn to be coded in the ZoLj area of ​​the FM maps, in order to reduce the throughput required for encoding the feature maps.In another embodiment, the ARM predictive neural network is used to produce the expected probability (pr) of the possible value of the current sample. Entropy coding is adapted to this probability (as is known for Huffman or arithmetic entropy coding).

[0184] At the end of the process, the current coded value Vcnde of the FM card; being processed is coded.

[0185] The [Fig. 10] is a logic diagram representing a method for decoding feature cards which can be implemented by the decoding device of [Fig. 2] and by the decoding process of [Fig. 7]

[0186] These steps constitute sub-steps of step F22 described previously in support of figure 7. They aim to decode a current value Vd of a point of a feature map p^yi of the first group being processed using values ​​from the neighborhood.

[0187] In a substep F221, a neighborhood vector Cdn is established, comprising values ​​close to the value Vdn. This step is similar to the previously described step E241 and the same embodiments apply. This neighborhood vector consists of a number C of values, or data, corresponding to neighborhood values ​​(for example, C=4, C=8, C=9 or C=11 in the examples of Figures 11 to 16) located in the same map and / or in a different map from the plurality M of maps pj\ / [(jj). These values, being in a causal neighborhood of the value Vdn, are known to the decoder.

[0188] According to a first embodiment, these values ​​are used to determine the context of an entropic decoder to decode the current value during a Step F224. This decoding is similar to that used in the encoder, for example, CAB AC. The use of the neighborhood to produce context information is similar to that chosen in the encoder. For example, one can count the number of non-zero neighbor values ​​and associate a context with each number. Alternatively, one can perform comparisons between several neighbor values ​​and associate a given context with an ordering configuration among the neighbor values, for example, by sorting the neighbor values ​​in ascending order and associating a context with each possible order.

[0189] In a second embodiment, the neighborhood is used to predict the current value from an autoregressive model during a step F222. In this embodiment, the past values ​​are constituted by the context, and the difference between the predicted variable and the actual value is decoded and then entropically dequantized during step F224.

[0190] In a third embodiment, as illustrated in Figures 11 to 16, a predictive neural network ARM' is used in step F223 to predict the statistical characteristics of the variable to be decoded. The neighborhood vector is applied as input to the ARM' network to provide an output prediction of the current value. In one embodiment, the ARM network behaves as a function fv as described in Figure 4, defined by a set of statistical parameters (mean, variance, median, etc.) used to entropically decode the current value. In another embodiment, the predictive neural network ARM' is used to produce the expected probability (pr) of the possible value of the current sample. Entropic decoding is adapted to this probability (as is known for Huffman or arithmetic entropic coding).The ARM network is identical to the ARM network if its encoding was performed without loss. At the end of the process, the current decoded value Vnde of the FMd- card being processed is decoded.

[0191] Figures 11 and 12 illustrate a method for encoding or decoding feature maps according to one of the embodiments

[0192] In this illustration, the encoding (respectively, decoding) of the current value Vn (respectively, Vdn) located at a current point with coordinates (xn, yn) in the characteristic area ZoLj of the current feature map FM (respectively, p^jj) uses the contextual information of this current map. This contextual information is represented by a neighborhood of the current point Vn. In the example shown here, the neighborhood used comprises 8 points belonging to a causal neighborhood of the current point associated with a 3x3 square mask.

[0193] This contextual information is, however, conditioned by the characteristic zone ZoLj in the current map. More precisely, the contextual information depends on the location of the current point relative to the boundary of the ZoLj zone. This location is obtained by evaluating a proximity criterion Cp(Vn) (resp. Cp(Vdn)) of the current point relative to a boundary of the characteristic zone. In the example described here, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is the Euclidean distance between the current point and the boundary of the ZoLj zone. Alternatively, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) can 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 between the current point and the boundary of the ZoLj zone.In yet another variant, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) can also be a quasi-distance or a gap or any other relevant measure indicating the proximity of the current point to the boundary of the ZoLj zone.

[0194] Two cases must then be distinguished: - In the first case, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is greater than a threshold S (for example predefined) and the values ​​of the neighborhood points in the current FM map; (resp. are used to encode (resp. decode) the current value Vn (resp. Vdn). - In the second case, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is less than this threshold S. In this case, the neighborhood vector Cn (resp. Cdn) is defined component by component. The value of a component of the neighborhood vector Cn (resp. Cdn) is equal to the value of the associated neighborhood point in the current map if that point belongs to the ZoLj zone, and to a replacement value otherwise.

[0195] Fig. 11 illustrates the first case while Fig. 12 illustrates the second case.

[0196] In [Fig.12], the values ​​of the points shown in grey are part of the area ZoLj and are used to determine components of the neighborhood vector Cn. The values ​​of the points shown in white do not belong to the ZoLj zone but to another ZoLx zone and cannot be used to define the neighborhood vector Cn, they are replaced by a replacement value Rp.

[0197] In one embodiment, such as that described in [Fig. 12], the replacement value is a constant, for example equal to the average value of the values ​​of the points in the ZoLj area in the current map or to a predetermined constant value.

[0198] In another embodiment shown in [Fig. 13], the replacement value is calculated based on the value of the neighborhood points belonging to the ZoLj zone

[0199] In another embodiment, a set of replacement values ​​is calculated based on the values ​​of neighborhood points belonging to the ZoLj zone. A neural network can be implemented to calculate a given number of replacement values ​​based on the neighborhood points of the ZoLj zone.

[0200] In the preceding embodiments, the definition of the neighborhood of the current point and the dimension of the neighborhood vector Cn are independent of the proximity criterion Cp(Vn) (resp. Cp(Vdn)). In another embodiment, this form of the neighborhood and / or the dimension of the neighborhood vector Cn may depend on the proximity criterion Cp(Vn) (resp. Cp(Vdn)).

[0201] Fig. 14 presents an example where the definition of the neighborhood is independent of the proximity criterion Cp(Vn) (resp. Cp(Vdn)) but where the dimension of the neighborhood vector Cn depends on this proximity criterion Cp(Vn) (resp. Cp(Vdn)).

[0202] Thus, when the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is greater than the threshold S, the neighborhood vector Cn is determined as explained with reference to [Fig.11].

[0203] On the other hand, when the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is less than the threshold S, the neighborhood vector Cn is determined as follows: - the number of points Np in the causal neighborhood of the current point belonging to the characteristic zone ZoLj is calculated and the neighborhood vector Vn is defined as a vector of dimension Np, and - the value of each point in the neighborhood of the current point belonging to the zone ZoLj is associated, according to an order for example predefined, with a different component of the neighborhood vector Cn.

[0204] Moreover, in the preceding embodiments, the contextual information represented by the neighborhood of the current point is independent of the proximity criterion Cpn. Alternatively, and as shown in [Fig. 15], the definition of the neighborhood of the current point may depend on the proximity criterion Cpn.

[0205] Two cases must again be distinguished: - In the first case, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is greater than the threshold S and the neighborhood values ​​of the current card FM; (resp. FM^i) belonging to a first neighborhood (identical to that defined previously in connection with [Fig.1 1]) are used to encode (resp. decode) the current value Vn (resp. Vdn). - In the second case, the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is less than this threshold S. In this case, a second neighborhood of the current point is used. In the example shown in [Fig. 15], the neighborhood used is a causal neighborhood comprising 10 points included in a 4x3 mask of the current point. The neighborhood vector Cn (resp. Cdn) of the current point is then defined component by component. The value of a component of the neighborhood vector Cn (resp. Cdn) is equal to the value of the associated point in the second neighborhood if that point belongs to the ZoLj zone, and to a replacement value otherwise.

[0206] Once determined, the neighborhood vector Cn is used to encode (resp. decode) the current value in one of the embodiments described previously in support of [Fig.9] (resp. [Fig. 10]).

[0207] Figure 16 illustrates another method of encoding or decoding feature maps according to one of the embodiments.

[0208] In this illustration, the encoding (resp. decoding) of the current value Vn (resp. Vdn) located at coordinates (xn, yn) in the current feature map FMi (i.e., the one being encoded or encoding simulated) (resp. "i.e., the one being decoded) uses the contextual information of its own map and the previous map FM2 (resp. FMdp-1). The values ​​located at coordinates (xn, yn-1), (xn, yn-2), (xn-1, yn-1), (xn-1, yn-1), (xn-1, yn), (xn-1, yn+1), (xn+1, yn-2), (xn+1, yn-1), (xn+1, yn), (xn+1, yn+1) in the FMi layer (resp. FMi) and the value located at coordinates (xn, yn) in the FMi layer are used. FM2(resp. are used to determine the neighborhood to encode (resp. decode) the current value.

[0209] This contextual information is conditioned by the ZoLj zone in the current map and in the previous map. The contextual information depends on the location of the current value Vn (resp. Vdn) relative to the boundary of the ZoLj zone in the current map and in the previous map.

[0210] The neighborhood vector is determined differently depending on the evaluation of the proximity criterion Cp(Vn) (resp. Cp(Vdn)). If the proximity criterion Cp(Vn) (resp. Cp(Vdn)) is greater than a threshold S, the neighborhood values ​​of the current map and the previous map are used to encode (resp. decode) the current value. Otherwise, the neighborhood vector Cn (resp. Cdn) of the value Vn (resp. Vdn) is defined component by component. The value of a component of the neighborhood vector Cn (resp. Cdn) is equal to the value of the associated point in the current map or the previous map if that point belongs to the ZoLj zone, and to a replacement value otherwise.

[0211] Figure 16 illustrates this latter case in the case of 9 points in a causal neighborhood included within a 3x3 mask in the current map and a one-pixel mask in the previous map. The values ​​shown in gray are part of the ZoLj area and are used to determine components of the neighborhood vector Cn. The values ​​shown in white do not belong to the ZoLj area but to the ZoLx area and, since they cannot be used to define the neighborhood vector Cn, are replaced by a replacement value Rp.

[0212] Once determined, the neighborhood vector Cn is used to encode (resp. decode) the current value in one of the embodiments described previously in support of [Fig.9] (resp. [Fig. 10]).

[0213] In the embodiments illustrated in Figures 11 to 16, the neighborhood vector C n is extracted by a CTX module (resp. CTX') from the FMC (resp. FMD) coding module, and then it is applied to the input of the prediction neural network ARM (resp. ARM'), used to predict the statistical characteristics (p,o) or the probability (pr) of the value to be coded (resp. decoded) by the entropy coder CE (resp. DE).

[0214] It should also be noted that the invention is not limited to the embodiments described above. It will indeed 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.

[0215] For example, NNSYN / NNSYN' synthesis neural networks can be recurrent neural networks.

[0216] In another example, the synthetic neural networks can consist of one or more convolutional neural networks, followed by an MLP, and then followed by one or more more convolutional neural networks, each of the convolutional neural networks being associated with a convolution kernel of predefined size, for example, 3x3. In these examples, the acquisition of the vectors Zn / Zdn is adapted to the topology of the synthetic neural networks NNSYN / NNSYN', i.e., several values ​​are extracted from at least one FM feature map. In other words, Zn is an L-tuple with L>M, constructed from values ​​taken from the FM maps at coordinates close to those of the current pixel, for example, on the support of the kernel of the convolution filter of one of the convolutional neural networks.

[0217] 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 to include all equivalents that are foreseeable by a person skilled in the art. applying his general knowledge to the implementation of the teaching that has just been disclosed to him.

Claims

1. Demands A method for encoding a signal (I(Pn)) comprising at least two zones, comprising the following steps for a zone to be coded (Zoj) of said signal comprising a plurality of samples (Pn) to be coded: - the construction (E22) of a group of latent value maps representative of the signal, - obtaining a segmentation (E24) of said group of latent value maps (FM;) into at least two characteristic areas, one of the at least two characteristic areas, called the associated characteristic area (ZoLj), being associated with said area to be coded, - for at least one sample, called the current sample (Pn), of the area to be coded, associated with a position (xn, yn) in the signal to be coded and with a point in the associated characteristic area (ZoLj), called the associated point: • the construction (E26, E27) of a characteristic vector (Zn) from said latent value maps (FM;) of said group, as a function of said position (xn, yn) of said current sample (Pn ), • the application (E28) of said characteristic vector (Z n) to the input of an artificial neural network, called a synthetic neural network (NNSYN), defined by a set of parameters (Wk), to obtain, at the output of said synthetic neural network, a vector (P'n) representative of a decoded value of the current sample (Pn), • the encoding of a value, called the current value (Vn) of said associated point, said encoding of the current value comprising the evaluation (F221) of a proximity criterion of said associated point with respect to a boundary of said characteristic zone, the construction (F221) of a neighborhood vector (Cn) of said associated point from a neighborhood of said associated point in said latent value maps of said group and the evaluation of said

2. proximity criterion, and an entropy coding (F223, F224) of said current value as a function of said neighborhood vector (Cn), and • the updating (E23, E26, E27) of at least one value of one of said latent value maps of said group and / or of at least one parameter of said synthetic neural network, as a function of a coding performance measure, - the coding of a group of latent value maps representing said area to be coded (Zo*) of said signal, and - the coding (E29) of said parameter set (Wk) of said synthesis neural network. A method for decoding a zone (Zodj), referred to as the zone to be decoded, of a signal comprising at least two zones, said zone to be decoded comprising a plurality of samples (Pdn) to be decoded, said decoding method comprising the following steps: the decoding of a group of latent value maps (FMd^) representative of said area to be decoded (Zodj) of said signal, said decoding comprising • the determination (F21) in said latent value maps (pMd-) of said group of a characteristic zone (ZoLj) corresponding to said zone to be decoded (Zodj) of said signal, • for at least one value, called current value (Vn), of a point of said characteristic zone (ZoLj), the evaluation of a proximity criterion of said point with respect to a boundary of said characteristic zone, the construction (F221) of a neighborhood vector (Cn) of said point from a neighborhood of said point in said latent value maps of said group and of the evaluation of said proximity criterion, and an entropic decoding (F224) of said current value as a function of said neighborhood vector (Cn), - the decoding (F26) of a set of parameters (Wdk) representative of a neural network (NNSYN'), called a synthetic neural network, and - the processing of said group of latent value maps (FMd^ P31 'c æscaLl of synthetic neurons.

3. A decoding method according to the preceding claim in which the neighborhood of said point is independent of said evaluation of the proximity criterion.

4. A decoding method according to claim 2, wherein the neighborhood of said point is selected based on said evaluation of the proximity criterion

5. A decoding method according to any one of claims 2 to 4 wherein, during the construction of the neighborhood vector, a component of the neighborhood vector 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 zone (ZoLj) and to a replacement value otherwise.

6. Decoding method according to the preceding claim wherein said replacement value is dependent on the points of said characteristic zone (ZoLj) or of the points associated with said neighborhood belonging to said characteristic zone (ZoLj).

7. Decoding method according to claim 5 wherein said replacement value does not depend on the points of said characteristic zone (ZoLj).

8. Decoding method according to any one of claims 2 to 4 wherein the construction of the neighborhood vector (Cn) comprises: - the identification of the points of said neighborhood belonging to said characteristic zone (ZoLj), said identified points, - the initialization of the number of components of said neighborhood vector (Cn) to the number of identified points, and - the assignment of the value of each identified point to a different component of said neighborhood vector (Cn).

9. A decoding method according to any one of claims 2 to 8 wherein said proximity criterion of said point with respect to a boundary of said characteristic zone is a distance.

10. A decoding method according to any one of claims 2 to 9 further comprising a step of obtaining a mask of said area to be decoded in said signal (I(Pn)) and in which said characteristic area (ZoLj) is determined from said mask.

11. A decoding method according to any one of claims 2 to 10 wherein said neighborhood vector (Cn) is used to determine an entropic decoding context of said current value (Vn).

12. A decoding method according to any one of claims 2 to 11 in which said entropic decoding of said current value (Vn) comprises a step of predicting said current value as a function of said neighborhood vector (Cn).

13. A decoding method according to any one of claims 2 to 12, characterized in that the decoding of said current value (Vn) of one of said latent value maps comprises: - the decoding of a set of parameters (Odk) representative of a neural network, called a prediction neural network (ARM'), and - the processing of said neighborhood vector (Cn) by the prediction neural network (ARM'), to provide a prediction of said current value (Vn).

14. A signal coding device (ENC) (I(Pn)) comprising at least two zones, characterized in that said coding device is configured to implement, for a zone to be coded (Zo1) of said signal comprising a plurality of samples (Pn) to be coded: - the construction (E22) of a group of latent value maps representative of the signal, - obtaining a segmentation (E24) of said group of latent value maps (FM;) into at least two characteristic zones, one of the at least two characteristic zones, called the associated characteristic zone (ZoLj), being associated with said zone to be coded, - for at least one sample, called the current sample (Pn), of the zone to be coded, associated with a position (xn, yn) in the signal to be coded and with a point of the associated characteristic zone (ZoLj), called the associated point:

15. • the construction (E26, E27) of a characteristic vector (Zn) from said latent value maps (FM;) of said group, as a function of said position (xn, yn) of said current sample (Pn ), • the application (E28) of said characteristic vector (Z n) to the input of an artificial neural network, called a synthetic neural network (NNSYN), defined by a set of parameters (Wk), to obtain, at the output of said synthetic neural network, a vector (P'n) representative of a decoded value of the current sample (Pn), • the encoding of a value, called the current value (Vn) of said associated point, said encoding of the current value comprising the evaluation (F221) of a proximity criterion of said associated point with respect to a boundary of said characteristic zone, the construction (F221) of a neighborhood vector (Cn) of said associated point from a neighborhood of said associated point in said latent value maps of said group and the evaluation of said proximity criterion, and an entropic encoding (F223, F224) of said current value as a function of said neighborhood vector (Cn), and • the updating (E23, E26, E27) of at least one value of one of said latent value maps of said group and / or of at least one parameter of said synthetic neural network, as a function of a coding performance measure, - the encoding of a group of latent value maps (pMçj) representative of said area to be coded (Zo1) of said signal, and - the coding (E29) of said parameter set (Wk) of said synthetic neural network. A decoding device for a zone (Zodj) of a signal comprising at least two zones, referred to as the zone to be decoded, said zone to be decoded comprising a plurality of samples (Pdn) to be decoded, characterized in that the encoding device is configured to implement: the decoding of a group of latent value maps (FMdp) representative of said zone to be decoded (Zodj) of said signal, said decoding comprising • the determination (F21) in said latent value maps (pMd-) of said group of a characteristic zone (ZoLj) corresponding to said zone to be decoded (Zodj) of said signal, • for at least one value, called current value (Vn), of a point of said characteristic zone (ZoLj), the evaluation of a proximity criterion of said point with respect to a boundary of said characteristic zone, the construction (F221) of a neighborhood vector (Cn) of said point from a neighborhood of said point in said latent value maps of said group and of the evaluation of said proximity criterion, and an entropic decoding (F224) of said current value as a function of said neighborhood vector (Cn), the decoding (F26) of a set of parameters (Wdk) representative of a neural network (NNSYN'), called a synthesis neural network, and the processing of said group of latent value maps (FMd-) P31 'c synthesis neural network.

16. 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 13 when said program is executed by a computer.

Images