WO2004112398A1

WO2004112398A1 - Method and device for encoding and decoding a video image sequence

Info

Publication number: WO2004112398A1
Application number: PCT/EP2004/006237
Authority: WO
Inventors: Sébastien BRANGOULO; Patrick Gioia; Nathalie Laurent
Original assignee: France Telecom
Priority date: 2003-06-12
Filing date: 2004-06-08
Publication date: 2004-12-23
Also published as: FR2856228A1

Abstract

The invention relates to a method for encoding a video image sequence, during which: the image sequence is split into video image sub-sequences; a meshing is carried out on the images of the video sequence, and; the coefficients are determined for each image starting from at least one transformation into wavelets. The invention is characterized in that the method comprises the steps of associating a portion of the meshing of an image to each video sub-sequence and of forming a data signal containing the portion of the associated meshing and of the information representative of the coefficients. The invention also relates to an encoding device, and to the associated method and device for decoding.

Description

METHOD AND DEVICE FOR ENCODING AND DECODING A SEQUENCE OF VIDEO IMAGES

The present invention relates to a method and a device for coding and decoding image sequences.

More precisely, the invention lies in the field of the adaptation over time of the coefficients obtained on a mesh of an image, the coefficients being produced by second generation wavelet transforms.

Several image coding techniques are known to date, such as for example discrete cosine transformation (DCT) techniques based on block structures, such as those proposed by the ISO / MPEG standardization bodies acronym for International Standardization Organization. / Moving Picture Expert Group and / or ITU-T acronym for International Télécommunication Union - Télécommunication Standardization Sector.

According to these video coding standards, a video sequence is processed so as to remove the spatial and temporal redundancies of the images of the sequence. The coded sequence, more particularly according to the MPEG-4 type coding, then consists of a set of I / B / P images. The I images, called Intra images are encoded in the same way as still images and serve as a reference for the other images in the sequence. The images P, called predicted images contain two types of information: error information compensated for movement and motion vectors representative of the movement relative to a previous image. These two pieces of information are deduced from one or more preceding images which can be of type I or of type P. The images B, called bidirectional images also contain these two pieces of information but are the latter are now based on two references, namely a reference to an image before image B during the reproduction of the video sequence and a reference to an image after image B during the reproduction of the video sequence, the anterior and posterior images can be of type I or type P.

Coding techniques using block structures lead to the appearance of strong effects, or artifacts, which greatly reduce the visual quality of the image reproduction. MPEG-4 or ITU-T / H263 encodings are now considered to have reached their limits, in particular because of the structure of rigid blocks of fixed size which are used as support for all the calculations and encoding operations. This is all the more true when the images and / or sequences of images are highly compressed.

Other image coding techniques instead use discrete cosine transforms wavelet transforms. Wavelet transforms are a very simple mathematical tool for breaking down functions in a hierarchical diagram. Briefly, the wavelet transforms make it possible to decompose a function into an approximation function and another function of details which influences the original function at various scales. The wavelet transforms are thus adapted to a multiresolution analysis of an image for example.

In the same way as coding techniques using discrete cosine transforms, coding techniques using wavelet transforms generate, when the images and / or image sequences are highly compressed, an over-oscillation or "ringing" effect. »In Anglo-Saxon terminology giving the rendered image a blurred aspect.

In order to reduce the amount of information transmitted, it is known not to transmit certain information of an image in a sequence of images if these are little different from that of the previous image. These techniques also create degradation of the reconstructed images. It has in fact been observed that with such a technique, the quality of the reconstructed images deteriorates over time.

The aim of the invention is to solve the problems of the prior art by proposing a method of coding a sequence of video images, the sequence of video images being broken down into sub-sequences of video images, a mesh being performed on the images of the video sequence, coefficients being determined for each image from at least one wavelet transform, characterized in that the method comprises the steps of:

- association of part of the mesh of an image with each video sub-sequence,

- formation of a data signal comprising the part of the associated mesh and information representative of the coefficients.

Correlatively, the invention proposes a device for coding a sequence of video images, the sequence of video images being broken down into sub-sequences of video images, a mesh being made on the images of the video sequence, coefficients being determined for each image from at least one wavelet transform, characterized in that the device comprises:

means for associating part of the mesh of an image with each video sub-sequence,

- Means for forming a data signal comprising the part of the associated mesh and information representative of the coefficients.

Thus, the data signal representative of the video image sequence is of reduced size without degrading the quality of the video image sequence when it is decoded. Associating a part of a mesh of an image with a sub-sequence of video images avoids having to transmit for each video image of the sequence of video images said part of the mesh.

More precisely, the part of the mesh is the topology of the mesh and the transformation of the images into coefficients is carried out by transforming the geometry of the meshes of each of the images of the video sub-sequence into coefficients from transforms into second generation wavelets.

Thus, by associating part of a mesh of an image and more precisely the topology of the mesh with a sub-sequence of video images, we avoid having to transmit for each video image of the sequence of video images the topology of its mesh. Thus, the images of the video sub-sequence varying little, the inclusion of the coefficients of each image of the sub-sequence with only one topology guarantees, when decoding the sequence of video images, good quality of reproduction of the sequence. decoded images. More precisely, the information representative of the transformed coefficients ′ is the coefficients of at least one reference image of the video image sub-sequence and for each other image of the video image sub-sequence, the information is representative of the difference between the coefficients of each other image and the coefficients of the reference image. Thus, by including in the data signal the difference between the coefficients of each other image and the coefficients of the reference image, the number of bits necessary for coding the sequence of video images is reduced. Since the video image sub-sequences are made up of a series of images exhibiting only few variations between them, the differences are then minimal and can therefore be represented on fewer bits.

Advantageously, a predetermined number of reference images is associated with a sub-sequence of video images as a function of the number of images contained in the sub-sequence of video images.

Thus, when a sub-sequence of images is relatively long, several reference images make it possible to form sub-parts of sub-sequences of video images. The differences between the coefficients in these subparts remain minimal and can therefore be represented on fewer bits.

Preferably, information is inserted into the data signal to differentiate the coefficients of the at least one reference image from the other coefficients.

Thus, the coefficients of the at least one reference image are easily identifiable during the decoding of the sequence of video images.

More precisely, the mesh is a regular mesh.

Advantageously, at least two different types of wavelet transforms are applied to at least two different regions of at least one image of the video sub-sequence.

In fact, the different types of wavelet transforms that exist have distinct coding properties. We exploit these different properties by applying to different regions of an image, the type of wavelet transform whose properties are best suited to the content of each of the regions of the image.

The overall coding of the video image sequence is thus optimized by adapting the coding by wavelet transforms to regions of the image with different characteristics, and by using, if necessary, several types of distinct wavelet transforms for coding. of the same image from the video sequence.

More precisely, the video image sub-sequences in the video image sequence are determined by comparison of differences in image coefficients of the video sequence at a predetermined threshold. Alternatively, the video image sub-sequences in the video image sequence are determined by comparing the signal-to-noise ratio between an image and a motion compensated image with respect to a predetermined threshold.

The invention also relates to a method of decoding a data signal representative of a sequence of video images, characterized in that the method comprises the steps of:

- determination in the data signal, of images of a sub-sequence of video images,

- obtaining the coefficients of the images of the sub-sequence of video images from information contained in the data signal, - transformation of the coefficients obtained according to an inverse transform of transform into wavelets,

- reconstruction of each of the images of the video sub-sequence from the transformed coefficients and of a part of the mesh associated with the sub-sequence of video images and contained in the data signal. Correlatively, the invention proposes a device for decoding a data signal representative of a sequence of video images, characterized in that the device comprises:

- means for determining in the data signal, images of a sub-sequence of video images, - means for transforming the coefficients of the images of the video sub-sequence obtained from the data signal according to a transform in wavelets, means of reconstruction of each of the images of the video sub-sequence from the transformed coefficients and of a part of the mesh associated with the sub-sequence of video images and contained in the data signal.

The invention also relates to a data signal representative of a sequence of video images, characterized in that the sequence of video images is divided into video sub-sequences, the signal comprising for each sub-sequence of video images information representative of the coefficients of the images of the sub-sequence of video images and information representative of part of a mesh of an image of the sequence of video images. More specifically, the information representative of the coefficients of the images of the sub-sequence of video images consist of the coefficients of at least one reference image of the sub-sequence of video images and for each other image of the sub-sequence of video images of the difference between the coefficients of each other image and the coefficients of the reference image and in that the signal comprises at least one predetermined piece of information identifying the coefficients of the at least one reference image of the sub -sequence of video images.

More precisely, the part of the mesh is the topology of the mesh of an image of the sequence of video images and the coefficients of the images of the sub-sequence of video images are coefficients of a transform into wavelets of second generation of the geometry of the meshes of each of the images of the sub-sequence of video images.

The advantages of the decoder and the data signal being identical to those mentioned for the coder, these will not be recalled.

The invention also relates to the computer program stored on an information medium, said program comprising instructions making it possible to implement the coding method and / or the decoding method described above, when they are loaded and executed by a computer system.

The characteristics of the invention mentioned above, as well as others, will appear more clearly on reading the following description of an exemplary embodiment, said description being made in relation to the accompanying drawings, among which: FIG. . 1 shows a device implementing the invention; Fig. 2 represents the algorithm for coding image sequences according to a first embodiment of the invention; Fig. 3 represents the algorithm for meshing an image and for determining a type of second generation wavelet transform to be applied to at least part of an image according to the invention; Fig. 4a represents the algorithm for determining, according to a second embodiment, video sub-sequences from the video image sequence as well as the reference images, in these video sub-sequences, used for coding the other images of their sub - respective video sequence, FIG. 4b represents the coding algorithm according to a second embodiment of the other video images of each video sub-sequence with respect to their respective reference image; Figs 5a and 5b show examples of image sequences in which sub-sequences of images are determined according to the first and second embodiments; Fig. 6 shows the decoding algorithm according to the invention of a data signal representative of a sequence of video images.

We will first describe the foundations of the so-called second generation wavelet transforms. These wavelet transforms were introduced by W. Dahmen in an article “Decomposition of refinable spaces and applications to operator equations” Numer. Algor., N ° 5 1993, pp 229-245, by J.M Carnicer, W Dahmen and JM. PENA in an article “Local decomposition of refinable spaces” Appl Comp. Harm. Anal. 3, 1996, pp 127-153 then developed by Wim Sweldens in a document entitled "The lifting Scheme: A construction of second generation wavelets", SIAM Journal of Mathematical analysis, Nov 1996.

These wavelet transforms are constructed from an irregular subdivision of the analysis space, and are based on a weighted and averaged interpolation method. These wavelets are particularly well suited for analyzes on compact supports and on intervals. They retain the properties of first generation wavelets, namely good time-frequency localization and good calculation speed. Second generation wavelets transforms have a number of properties. Like the first generation wavelet transforms, they form a base of Riez on L (R) as well as a uniform base for a wide variety of function spaces. Like the first generation wavelet transforms, the coefficients of decomposition on the uniform basis are known or can simply be determined. Wavelet transforms are either orthogonal or bi-orthogonal. Like the first generation wavelet transforms, the wavelet transforms and their dual models have local properties in space and frequency. The frequency localization properties result directly from the regularity of the wavelet transform for the high frequencies and from the number of zero polynomial moments for the low frequencies. Like the first generation wavelet transforms, the second generation wavelet transforms can also be used in multiresolution analysis. Second-wavelet transforms also have fundamental properties that first-generation wavelets do not have.

Second generation wavelet transforms can be defined on arbitrary domains, such as curves, surfaces or manifolds. The transforms into second generation wavelets allow analysis on curves and surfaces.

Second generation wavelet transforms allow the use of algorithms suitable for irregular data samples.

The multiresolution analysis and the application of these second generation wavelet transforms over a bounded interval, and in particular on a triangular mesh are presented in an article by M Lounsbery, T DeRose, J Warren entitled “Multiresolution Analysis for Surfaces of Arbitrary Topological Type ”ACM Transactions on Graphics, 1994. This multiresolution analysis is intimately linked to the recursive subdivision process. The surfaces of three-dimensional objects or images of a video sequence are represented by a polyhedral mesh. The mesh is a linear surface in pieces, consisting for example of triangular faces. The representation of the surfaces by a triangular mesh makes it possible to compress, edit, efficiently transfer sequences of video images. When one describes a mesh, and more particularly a triangular mesh, this one is made up of two parts. A first part representing the connectivity of the vertices, edges and faces determining the topology of the mesh and a second part comprising all the positions of the vertices thus defining the geometry of the mesh. The document by Wim Sweldens, entitled "The lifting Scheme: A construction of second generation wavelets", SIAM Journal of Mathematical analysis, Volume 29, number 2, pp 511-546, 1998 describes a generalized transformation method for calculating the coefficients of these transformed into wavelets without having to use convolution products as well as Fourier transforms. Thus, it is an excellent tool for the construction of second generation wavelet transforms in the sense that the Fourier transformation is no longer necessary. This method allows a reconstruction, by simple operations on lines and columns of an analysis matrix to separate the signal to be processed into even and odd samples, to predict the odd samples according to the even samples. Once the prediction has been made, the signal is updated in order to preserve the initial properties. This algorithm can be repeated as many times as necessary.

We will now describe with reference to FIG. 1 a coding and / or decoding device implementing the invention. This device 10 is suitable for coding a digital signal consisting of a video sequence and / or for decoding a digital signal consisting of a video sequence previously coded according to the invention.

The device 10 is for example a microcomputer. It can also be a means of viewing video image sequences such as a television set or any other device for coding and / or decoding video image sequences. The device 10 includes a communication bus to which a central unit is connected

100, a read-only memory 102, a random access memory 103, a screen 104, a keyboard 114, a hard disk 108, a CD / CD player / recorder, a communication interface with a communication network 113, a memory card. input output 106 connected to an image sequence capture means 107 such as a camera.

The hard disk 108 stores the programs implementing the invention, as well as the data processed according to the invention. These programs can also be read via the compact disc or received via the communication network 113, or even memorized in read-only memory 102. More generally, the programs according to the present invention are memorized in a storage means. This storage means can be read by a computer or a microprocessor 100. This storage means is integrated or not into the device, and can be removable. When the device 10 is powered up, the programs according to the present invention are transferred into the random access memory 103 which then contains the executable code of the invention as well as the variables necessary for the implementation of the invention. The device 10 can receive data to be processed from a peripheral device 107, such as a digital camcorder, or any other means of acquiring or storing data.

The device 10 can also receive data to be processed from a remote device via the communication network 113 and / or transmit a data signal comprising the data coded according to the invention to a remote device via the same communication network 113.

The communication network 113 is for example an Internet type network or a telephone telecommunication network by which a videoconference type communication is established between two devices 10. The communication network 113 can also be a Hertzian or satellite broadcasting network. video information encoded according to the present invention.

The device 10 includes a screen 104 capable of reproducing the sequences of video images decoded according to the invention and / or the sequences of video images to be processed according to the invention. The device 10 also includes a keyboard 114 serving as a user interface.

Via this keyboard 114, the user can select certain parameters of the invention and / or activate the programs according to the invention.

Fig. 2 represents the algorithm for coding video image sequences according to a first embodiment of the invention. When the application is launched, the processor 100 of the device 10 reads from the read-only memory 102, the program instructions corresponding to steps E200 to E214 in FIG. 2 and loads them into RAM 103 to execute them.

In step E200, the processor 100 takes the first image II of a video sequence I to be coded and determines a mesh of the latter. A mesh is a graph representation of the image. The determination of the mesh will be explained later in more detail with reference to FIG. 3.

The video sequence is for example stored on a CD disc included in the player 109 or received via the input / output interface 106 of an image capture device 107 or also received from the communication network 113 by via the communication interface 112. The video sequence I is for example the sequence of video images shown in FIG. 5a. This sequence of video images I is in our example made up of seventeen images denoted II to 117.

The mesh thus determined, the processor 100 in step E201 determines the coefficients of transformation into second generation wavelets from the mesh. For this, we apply the wavelet transform using the Lifting technique on all the positions of the vertices defining the geometry of the mesh. The second generation wavelet transform is performed for a predetermined number of resolution levels. The lifting technique is similar to that described in the document by Wim Sweldens, entitled "The lifting Scheme: A construction of second generation wavelets".

It should be noted that, in a particular mode, as will be described with reference to FIG. 3, different types of wavelet transforms are applied to different parts of the image to be processed. For example, a wavelet transform known as Butterfly wavelets is assigned to the texture areas, a wavelet transform known as Loop wavelets is assigned to the contours of natural objects, a wavelet transform known as Catmull-Clark wavelets is assigned to the contours of unnatural objects and finally a wavelet transform known as affine wavelets is assigned to singularities.

This operation carried out, the processor 100 stores in step E202 the coefficients of the wavelet transform of the first image II in the random access memory 103. These coefficients are then considered as reference coefficients. This image II in FIG. 5a is marked with a letter R representative of its reference image function.

The processor 100 in the next step E203 takes the next image 12 of the video sequence I to be coded and determines a mesh of the latter. The determination of a mesh will be explained later with reference to FIG. 3. The mesh thus determined, the processor 100 in step E204 determines the coefficients of the second generation wavelet transform from the mesh. For this, we apply the wavelet transform for a predetermined number of resolution levels using the Lifting technique on the set of vertex positions defining the geometry of the mesh of the image being processed.

It should be noted that, as previously, different types of wavelet transforms are applied to different parts of the image to be processed. This operation performed, the processor 100 reads in step E205 the reference coefficients stored in step E202 of the wavelet transform of the first image II in the random access memory 103.

The processor 100 then performs in step E206 the difference between the reference coefficients of image II and the coefficients of the second generation wavelet transform determined for image 12 in step E204. Of course, the difference is calculated for each coefficient corresponding to an identical or similar vertex of the same surface in the two images II and 12.

The calculated difference, the processor 100 checks in step E207 if the calculated differences are less than a predetermined threshold. The difference between each coefficient can be used for this verification or an average of the differences can also be used. This threshold is either predetermined and fixed, or modifiable by the user via the keyboard 114 of the device 10.

If so, the processor 100 goes to step E207 and checks whether part of the mesh of the second image 12 has moved with respect to the mesh of the reference image II. This corresponds to a movement of an object in the video sequence. If so, the processor 100 codes this movement and stores it in step E210. It should be noted that steps E209 and E210 may also not be performed by the processor 100 in the algorithm as described.

If a small displacement exists or if no part of the mesh of the second image 12 has moved with respect to the mesh of the reference image II, the processor 100 stores in step E211 the differences in coefficients calculated at step E206 in the random access memory 103.

This operation carried out, the processor 100 checks whether there are other images in the video sequence I to be coded. In our example, seventeen images are to be coded and the processor then returns to step E203.

The processor 100 performs in the same manner as that previously described the steps E203, E204, E205, E206, E207, E209, E210, E211 and E212.

It is the same for images 13 to 15. The differences of the coefficients between each of these images with the reference coefficients are thus memorized. The processor 100 also performs the same operations for the image 16 which, in our example, corresponds to a change of plane. The difference between the coefficients of the second generation wavelet transform of the images II to 16 being greater than the threshold noted Threshold at step E207, the processor 100 goes to step E208.

The processor 100 in step E208 checks whether other reference coefficients exist in the random access memory 103.

In our case, only the reference coefficients of the second generation wavelet transform of image II are stored. The processor 100 then goes to step E202 and stores in this step the coefficients of the second generation wavelet transform of the image 16 determined in step E204.

The reference coefficients used subsequently correspond to the coefficients of image 16.

The processor 100 performs in the same manner as that previously described the steps E203, E204, E205, E206, E207, E209, E210, E211 and E212 for the images 17 to 19.

In our example, image 110 corresponds to a change of plane. The difference between the coefficients of the second generation wavelet transform of the images 110 and 16 being greater than the threshold at step E207, the processor 100 goes to step E208.

In our example, the reference coefficients of image II are also stored in the random access memory 103. The processor goes to step E205, reads them and performs in step E206 the difference between the reference coefficients of the image

II and the coefficients of the second generation wavelet transform determined for image 110 in step E204.

The calculated difference, the processor 100 checks in step E207 if the calculated differences are less than a predetermined threshold. The advantage of keeping in memory the coefficients of the different reference images, in this case II and 16, is based on the fact that if the video sequence returns to a previous scene in a relatively short time, the stored reference coefficients do not do not need to be recalculated, which saves a considerable amount of time. Similarly, when inserting the reference coefficients into the data signal, these will only be inserted once. A simple indicator in the data signal will indicate which coefficients in the data signal should be considered as reference coefficients for each video sub-sequence. According to our example, image 110 corresponds to a new plane in the video sequence. The test of step E207 is therefore negative. The processor then returns to step E205.

The processor 100 performs in the same manner as that previously described the steps E203, E204, E205, E206, E207, E209, E210, E211 and E212 for the images 111 to I17.

Thus, according to the example of FIG. 5a, the processor 100 has stored in random access memory 103 the coefficients of the wavelet transform of the images II, 16 and 110 as well as a part of their respective meshes. This part is representative of the connectivity of the vertices, the edges and the faces determining the topology of the mesh.

Still according to the example of FIG. 5a, the processor 100 has stored in RAM 103 the differences between the coefficients of the wavelet transform of images 12 to 15 and the coefficients of the wavelet transform of image II, the differences between the coefficients of the wavelet transform of images 17 to 19 and the coefficients of the wavelet transform of image 16 and the differences between the coefficients of the wavelet transform of images II 1 to 117 and the coefficients of the wavelet transform of image 110.

Thus, the video sequence of FIG. 5a has been subdivided into three video sub-sequences denoted S1 to S3. Each video sub-sequence is associated with at least part of a mesh and the coefficients of the second generation wavelet transform of a reference image in the video sub-sequence, the difference between the coefficients of the second generation wavelets of each other image of the video sub-sequence and the coefficients of the wavelet transform of the reference image. It should be noted that, as a variant, the difference can be calculated between the coefficients of the second generation wavelet transform of an image of the video sub-sequence I and the coefficients of the wavelet transform of the following image. All the images II to 117 having been processed, the processor 100 in step E212 goes to step E213.

In step E213, the processor 100 codes the coefficients as well as the differences in coefficients obtained previously. These coefficients and their differences are for example preferably coded according to a Zerotree type technique. This technique is for example described in the publication "Embedded Image Coding Using Zerotree of Wavelet Coefficients" by J.M Shapiro IEEE Transcations on signal Processing, Vol 41, NO 12, December 1993.

Indeed, in the wavelet transform, each coefficient of a sub-band has four coefficients in the lower sub-band. Thanks to this structure, if a coefficient at a given low resolution sub-band is insignificant compared to a predetermined threshold, all the coefficients of the same orientation in the same spatial position at higher frequency sub-bands are also insignificant vis-à-vis this same predetermined threshold. A null symbol tree is then defined starting at a root which is also null and referenced as an end of block. Thus, many coefficients at higher frequency sub-bands can be deleted. This algorithm is interesting in the sense that coding can stop at any point. It also allows precise encoding in relation to a given bit rate.

Alternatively, coding can also be done using the EBCOT (Embedded Block Coding with Optimized Truncation of the embedded bit-stream) type method. This method is described in the publication by D Taubman "High performance scalable Image compression with EBCOT" IEEE Transactions on Image

Processing Vol 9, NO 7, July 2000.

Once this has been done, the processor 100 goes to step E214 and puts the coded coefficients in the form of a data or bitstream signal.

The coded coefficients are for example transmitted in order of priority. The data is transmitted in the form: packet number / header including inter alia the number of coefficients in the packet, information representative of the area of the image concerned, the number of bit planes used to code the image , the type of wavelet transform used to code certain areas of the image, information on the mesh such as the topology thereof as well as predetermined fields indicating that the coefficients according to these predetermined fields are image coefficients of reference according to the invention. Other fields indicate also the reference image used when calculating the differences in coefficients of the other images of the video sub-sequences.

Preferably, the signal is in the form of consecutive packets, each packet header comprising a packet start field, the packet number in question, the type of wavelet transform used, a reference image field making it possible to indicate that the coefficients of the packet are coefficients of reference images and are therefore not differences of coefficients between two images, a buffer number field indicating, when a sequence or a sub-sequence comprises a plurality of reference images, the image of reference having been used during the coding of the coefficients included in the subsequence.

The package number field contains a package identifier assigned in order according to the size of the package.

The information header field includes the following sub-fields:

- the number of coefficients in the package; - the zone of the image considered represented from the topology of the mesh;

- the number of bit plans used for coding the coefficients in question.

The standard wavelet transform field indicates whether the wavelet transform applied to the image area from which the packet comes is for example a wavelet transform of Loop, Butterfly, Catmull-Clark or even a wavelet transform of affine type, or any other type chosen according to the area of the image considered.

The mesh field is used to transmit the mesh topology of each reference image.

It should be noted that, as a variant, the data signal includes information indicating the reference image used for coding at least one other image, the reference image being part of a group of reference images encoded in the data signal. Fig. 3 represents the algorithm for meshing an image and for determining a type of second generation wavelet transform to be applied to at least part of an image according to the invention. When the application is launched, the processor 100 of the device 10 reads from the read-only memory 102, the program instructions corresponding to steps E300 to E308 of FIG. 3 and loads them into RAM 103 to execute them.

The mesh as described in the present algorithm is performed for each of the images to be coded.

In step E300, the processor 100 performs a regular and dense mesh of the image, that is to say that the image is subdivided for example into triangular surfaces. The density of this mesh is predetermined. It can also be adapted according to the image sequence or the image to be coded. Once this step has been completed, the processor 100 performs the loop consisting of the steps

E301 to E305. This loop corresponds to the realization from the regular and dense mesh of an optimal mesh by fusion of the triangles of the mesh on predetermined distortion rate criteria and according to the different properties of the different areas of the image. This optimal mesh gives an irregular subdivision of the image and nevertheless allows the use of second generation wavelet transforms on it.

The processor 100 in step E301 thus optimizes the triangles of the dense mesh, merges the neighboring triangles as a function of predetermined distortion rate criteria and permutes the edges of the merged triangles of the mesh. This determination of the mesh is according to a particular mode in accordance with that as described in the article by M Lounsbery, T DeRose, J Warren entitled "Multiresolution Analysis for Surfaces of Arbitrary Topological Type".

In the next step E302, the processor 100 quantifies the nodes of the new triangles formed. This operation carried out, the processor 100 goes to the next step E303 which consists in verifying that a mesh has not returned during the optimization step E301.

In the next step E304, the processor 100 quantifies the nodes of the new triangles formed. The processor 100 then checks in step E305 whether the loop made up of steps E301 to 305 has been carried out a predetermined number of times and, if not, reiterates this loop again.

It should be noted here that, as a variant, the determination of the mesh can be carried out in the opposite manner to that previously described. A first mesh coarse can be applied on the image, then, one divides the meshes into four meshes, for example until obtaining a semi-regular final mesh. This method is more particularly described in the publication by P Gioia "Reducing the number of wavelet coefficients by géometry partioning" Computational geometry, Theory and Applications Vol 14, 1999, pp 25-48.

If so, the processor 100 goes to step E306 which consists of carrying out edge management operations. The management of the edges consists in the application of a homeomorphism of the mesh to a torus according to the so-called periodization method. As a variant, the management of the edges consists in performing a symmetry of the data by extending the diagonals situated at the borders and which are not oriented in one of the directions of the mesh.

Thus, a semi-adaptive hierarchical mesh is produced. This step carried out, the processor 100 determines in the image in step E307 zones of distinct nature. These areas are non-limiting areas of singularity, areas of texture, areas of contours of natural or unnatural objects.

These zones are determined from the characteristics of the mesh and more particularly from the density of the latter around a point and of a region around this point. If at a point of the mesh, the density of mesh is important and around this one, the mesh is empty. This point is then considered as an isolated singularity. If at a point of the mesh, the density of mesh is important and around this one, the mesh is also dense. This area is then considered to be a texture area. If at a point of the mesh the density of mesh is important and around it and the density of mesh is strong according to a particular direction, this zone is then regarded as a zone of contour. The differentiation of a natural contour from an unnatural contour is carried out by analyzing the particular direction of the mesh. Indeed, natural objects have more uncertain contours than the contours of non-natural objects. The direction of unnatural contours is therefore more uniform than that of unnatural contours. The determination of the zones being carried out, the processor 100 goes to step

E308 and assigns to each specific zone a suitable type of wavelet transform.

For example, a wavelet transform known as Butterfly wavelets is assigned to the texture areas, a wavelet transform known as Loop wavelets is assigned to the texture areas. contours of natural objects, a wavelet transform known as Catmull-Clark wavelets is assigned to the contours of unnatural objects and finally a wavelet transform known as affine wavelets is assigned to singularities . The algorithm of FIG. 4a determines, according to a second embodiment, the video sub-sequences among the video sequence, as well as the reference images in these video sub-sequences used to encode the other images of their respective video sub-sequences.

In this second embodiment, the video sequence is not subdivided into video sub-sequences as a function of a comparison of differences in transform coefficients in second generation wavelets at a predetermined threshold. In this second mode, the video sequence is subdivided into video sub-sequences as a function of the comparison of PSNR (signal to noise ratio) between a compensated image in movement and the original image with respect to a predetermined threshold S. At the launch of the application, the processor 100 of the device 10 reads from the read-only memory 102, the program instructions corresponding to steps E400 to E421 of FIG. 4a and loads them into RAM 103 to execute them.

In step E400, the processor 100 initializes the variables i and k to the value 1. The variable i is a variable associated with the current image. The variable k is a variable associated with the subsequences of the sequence of images.

The processor 100, in the next step E401, reads the first image 121 of the video sequence. The video sequence is for example stored on a CD disc included in the player 109 or received via the input / output interface

106 of an image capture device 107 or also received from the communication network 113 via the communication interface 112. The video sequence is for example the video sequence shown in FIG. 5b. This video sequence is in our example made up of seventeen images denoted 121 to 137.

In step E402, the processor 100 assigns to the variable DEB (Ic) the value of the variable i. The variable Deb (k) corresponds to a variable representative of the start of a video sub-sequence.

The processor 100 then goes to step E404 which consists in reading the next image of the video sequence, in this case the image 122.

The processor 100 then determines in step E405 the movement between the previous image 121 and the image 122. This determination of movement is for example performed by performing block matching, or block matching, between the two images.

Then, in step E406, the processor 100 calculates the motion compensated image of the current image 122. Steps E405 and E406 preferentially use the motion compensation method as described in the document Joint final Committee Draft of Joint Video Specification (ITU-T Rec H.264 ISO / IEC 14496-10 AVC) Thomas Wiegand, Klagenfurt, July 22, 2002.

The processor 100 then goes to step E407 which consists in calculating the PSNR between the image compensated in movement and the current image 122.

This calculation carried out, in step E408, the processor 100 compares the PSNR previously calculated with a predetermined threshold S.

If the PSNR is greater than the threshold S, the image 122 is part of the same sequence as the image 121, which in our example is the case. There is little movement between the images. No plan change in the video sequence has taken place. The processor 100 then goes to step E409.

The processor 100, at this stage, assigns the variable End (k) representative of the end of a video sub-sequence to the current value of i, that is to say the value 2.

Once this step has been completed, the processor checks in step E410 whether the last image of the video sequence has been processed, which is not the case in our example. The processor 100 then returns to step E403.

The processor 100 will reiterate the loop made up of steps E403 to E410 as long as the calculated PSNR is greater than the threshold S.

In our example, the loop made up of steps E403 to E410 is repeated for the images 123 to 126.

When the processor 100 compares the PSNR calculated between the image compensated in movement and the current image 126, the latter is according to our example below the threshold S. The processor 100 therefore goes to step E411.

In step E411, the processor 100 assigns to the variable End (k) representative of the end of a video sub-sequence the current value of i, that is to say the value 6. The first video sub-sequence S21 thus consists of the images 121 to 125.

This step carried out, the processor checks in step E412 if the last image of the video sequence has been processed, which is not the case in our example. The processor 100 goes to step E413, increments the variable k by one unit and then returns to step E402.

In step E402, the processor 100 then assigns to the variable DEB (k) the value of the variable i. The processor 100 then goes to step E404 which consists in reading the next image of the video sequence, in this case the image 127.

The processor 100 then determines in step E405 the movement between the previous image 126 and the image 127.

Then, in step E406, the processor 100 calculates the motion-compensated image of the current image 127.

The processor 100 then goes to step E407 which consists in calculating the PSNR between the image compensated in movement and the current image 127.

This calculation carried out, in step E408, the processor 100 compares the PSNR previously calculated with a predetermined threshold S. If the PSNR is greater than the threshold S, the image 127 is part of the same sequence as the image 126, which in our example is the case. There is little movement between the images. No plan change in the video sequence has taken place. The processor 100 then goes to step E409.

The processor 100, at this stage, assigns the variable End (k) representative of the end of a video sub-sequence to the current value of i, that is to say the value 7.

The processor 100 will reiterate the loop made up of steps E403 to E410 as long as the PSNR is greater than the threshold S.

In our example, the loop made up of steps E403 to E410 is repeated for the images 126 to 129. When the processor 100 compares the PSNR calculated between the image compensated in movement and the current image 130, the latter according to our example is lower than the threshold S. The processor 100 therefore passes to step E411.

In step E411, the processor 100 assigns to the variable End (k) representative of the end of the second video sub-sequence the current value of i, that is to say the value 10. The second video sub-sequence S22 thus consists of images 126 to 129.

This step carried out, the processor checks in step E412 if the last image of the video sequence has been processed, which is not the case in our example. The processor 100 goes to step E413, increments the variable k by one unit and then returns to step E402. The processor 100 performs the same operations for the images 131 to 137 as those previously described. They will not be re-described.

The last video sub-sequence then consists of steps 130 to 137.

When the processor 100 has processed the image 137, the test of step E410 turns out to be positive. The processor therefore proceeds to the next step E414.

The processor 100 at this stage initializes the variable k to the value 1.

This operation performed, the processor 100 determines in step E415 the size denoted Size (k) of the current sub-sequence, in this case the sub-sequence S21. The size is determined from the End (k) and Deb (k) variables previously determined.

The processor 100 then compares in step E416 the size determined with a predetermined value T. Indeed, according to the invention, two different processing operations are carried out according to the size of the sub-sequence processed.

In our example, the size of the sequence S21 is equal to five images. This being less than T, the processor goes to step E417.

At this stage, the processor 100 determines the reference image of the subsequence. The reference image is here taken as the image placed in the middle of the video sub-sequence. This is particularly interesting because the variations between the different images of the sub-sequence are thus reduced. According to this embodiment, the difference between the images of the sub-sequence and the reference image is less. Fewer bits are then required to code these differences.

According to our example, image 123 is then the reference image.

Once this has been done, the processor 100 checks in step E420 whether all of the previously determined subsequences have been processed. According to our example, the test is negative and the processor 100 returns to step

E415.

The processor 100 similarly performs steps E415 to 417 for the second sub-sequence S22, and the image 128 is then the reference image.

For the sub-sequence S23, the processor 100 determines a size of eight images and considers this size as greater than the threshold T. The test of step E416 is therefore negative. The processor 100 therefore goes to step E418.

In step E418, the processor 100 determines the number M of reference images necessary for optimal coding of the sub-sequence. This number M is determined as a function of the number of images contained in the sub-sequence and by example, from a table stored in the read-only memory 102. By way of example, for eight images, M is considered to be equal to two. It should be noted that for reasons of simplification, a limited number of images is considered in the video sequence and the video sub-sequence. Of course, in a practical embodiment, the number of images in a video sub-sequence is much greater and a reference image is determined for a much greater number of images.

The number of reference images for a sub-sequence having been determined, the processor 100 goes to step E419 and determines the reference images in the sub-sequence S23 by cutting the sub-sequence S23 into M sub-parts so equidistant and chooses the central image of each sub-part as the reference image of the sub-part. According to our example, the sub-sequence S23 is formed of the two sub-parts denoted S'23 and S "23 and the image 131 is the reference image of the sub-part S'23 while the image 136 is the reference image of subpart S "23.

Once this operation has been carried out, the processor 100 checks whether all the sub-sequences have been processed and then proceeds to step E500 of FIG. 4b.

The algorithm of FIG. 4b codes according to a second embodiment the other video images of each video sub-sequence with respect to their respective reference image.

When the application is launched, the processor 100 of the device 10 reads from the read-only memory 102, the program instructions corresponding to steps E500 to E523 of FIG. 4b and loads them into RAM 103 to execute them.

The processor 100 initializes the variable k representative of the sub-sequence to be processed at the value 1.

The processor checks in the next step E501 if the current subsequence comprises several reference images. According to our example, the sub-sequence S1 has only one reference image, image 123. The processor then goes to step E502. At this stage, the processor 100 reads the reference image of the sub-sequence being processed, in this case the image 123.

The processor 100 then performs in step E503 a mesh on the reference image and the determination of the type of wavelet transform to be applied to areas of the image according to the algorithm of FIG. 3 previously described. This operation carried out, the processor 100 determines in step E504 the coefficients of the second generation wavelet transform from the mesh determined in the previous step. For this, we apply the wavelet transform, for a predetermined number of resolution levels, using the Lifting technique on all the positions of the vertices defining the geometry of the mesh of the image being processed.

It should be noted that, as before, different types of wavelet transform are applied to different parts of the image to be processed.

For example, a wavelet transform known as Butterfly wavelets is assigned to texture areas, a wavelet transform known as Loop wavelets is assigned to natural object contours, a wavelet transform known as Catmull-Clark wavelets is assigned to the contours of unnatural objects and finally a wavelet transform known as affine wavelets is assigned to singularities.

This operation carried out, the processor 100 checks in step E505 if there are other images in the sub-sequence being processed. We check more precisely if the subsequence consists of a single image. According to our example, the subsequence consists of a plurality of images. The processor 100 then goes to step E506.

In step E506, the processor 100 determines the position of the next image to be processed in the sub-sequence. This image is for example the neighboring image preceding the reference image of the sub-sequence.

Once this operation has been carried out, the processor 100 reads in step E507 the image which is at the position previously determined, in this case the image 122.

The processor 100 then performs in step E508 a mesh on the reference image in accordance with the algorithm of FIG. 3 previously described.

Once this has been done, the processor 100 determines in step E509 the second generation wavelet transform coefficients from the mesh determined in the previous step. For this, we apply the wavelet transform, for a predetermined number of resolution levels, using the Lifting technique on all the positions of the vertices defining the geometry of the mesh of the image being processed. It should be noted that, as before, different types of wavelet transform are applied to different parts of the image to be processed.

It should be noted here that the images of a sub-sequence are similar. The determination of a wavelet transform type for the images other than the reference image of a video sub-sequence may, according to a variant, not be carried out. The areas determined for the reference image are then considered to be identical for the other images.

The processor 100 performs in step E510 the difference between the coefficients of the second generation wavelet transform determined for image 122 in step E509 and the reference coefficients of image 123. Of course, the difference is calculated for each coefficient corresponding to an identical or similar vertex of the same surface in the two images 123 and 122. The difference of the coefficients for the image 122 is stored in the random access memory 103.

This operation performed, the processor 100 checks in step E511 whether the position of the image of the sub-sequence being processed corresponds to the position of the image of the start of the sub-sequence. According to our example in FIG. 5b, the test is negative. The processor 100 decrements the value of the variable Pos by one unit in step E512.

The processor 100 then performs steps E507 to E510 for the image 121 in the same manner as that previously described for the image 122.

The image 121 corresponding to the first image of the sub-sequence S21, the processor 100 passes from step E511 to step E513.

In step E513, the processor 100 determines the position of the next image to be processed in the sub-sequence. This image is for example the next neighboring image 124 of the reference image of the video sub-sequence.

Once this has been done, the processor 100 reads, at step E514, the image which is at the position previously determined, in this case the image 124.

The processor 100 then performs in step E515 a mesh on the reference image in accordance with the algorithm of FIG. 3 previously described. This operation carried out, the processor 100 determines in step E516 the coefficients of the second generation wavelet transform from the mesh determined in the previous step.

It should be noted that, as before, different types of wavelet transform are applied to different parts of the image to be processed. The processor 100 performs in step E517 the difference between the reference coefficients of image 123 and the coefficients of the second generation wavelet transform determined for image 124 in step E516. The difference of the coefficients for the image 124 is memorized in the random access memory 103. This operation carried out, the processor 100 checks in step E518 if the position of the image of the sub-sequence being processed corresponds to the position the end frame of the subsequence. According to our example in FIG. 5b, the test is negative. The processor 100 increments the value of the variable Pos by one unit in step E519. The processor 100 then performs steps E514 to E517 for the image 125 in the same manner as that previously described for the image 124.

The image 125 corresponding to the last image of the sub-sequence S21, the processor 100 passes from step E518 to step E520.

The processor 100 checks at this stage if other video sub-sequences in the video sequence are to be processed.

According to our example in FIG. 5b, two subsequences S22 and S23 must be processed. The processor 100 then goes to step E523 and then increments the variable k by one unit.

Sub-sequence S22 is then processed. The processor 100 returns to step E501 and performs steps E501 to E520 in the same manner as that previously described.

When the processor 100 has processed all of the images of the sub-sequence S22, it processes, according to our example, the sub-sequence S23.

It should be noted that the sub-sequence S23 comprises more than one reference image. The test of step E501 is therefore positive. The processor 100 then performs the following operations, not shown.

The processor 100 takes the reference image 131 of the sub-part S ′ 23 of the sub-sequence S23, performs a mesh of the latter and determines the coefficients of the second generation wavelet transform of the latter. The processor 100 reads each of the other images 130, 132 and 133 of the sub-part

S'23 of the sub-sequence S23, meshes them, determines the coefficients of the second generation wavelet transform of these and performs the difference between the coefficients of the reference image of the sub- part S'23 of the sub-sequence S23 and their respective coefficients. The processor 100 takes the reference image 136 of the sub-part S "23 of the sub-sequence S23, performs a mesh thereof and determines the coefficients of the second generation wavelet transform of the latter.

The processor 100 takes each of the other images 134, 135 and 137 of the sub-part S "23 of the sub-sequence S23, performs a mesh of these, determines the coefficients of the second generation wavelet transform of these. ci and make the difference between the coefficients of the reference image of the subpart

S "23 of the subsequence S23 and their respective coefficients.

Once these operations have been carried out, all the sub-sequences have been processed. The processor 100 goes to step E521.

In step E521, the processor 100 codes the coefficients as well as the differences in coefficients obtained previously. These coefficients and their differences are for example coded according to a Zerotree type technique previously described or the coding can also be carried out using the EBCOT type method. Once this has been done, the processor 100 goes to step E522 and puts the coded coefficients in the form of a data or bitstream signal.

The coded coefficients are for example transmitted in order of priority. The data is transmitted in the form: packet number / header including inter alia the number of coefficients in the packet, information representative of the area of the image concerned, the number of bit planes used to code the image , the type of wavelet transform used to code certain areas of the image, information on the mesh such as the topology thereof as well as flags indicating that the coefficients contained in the packet comprising a flag or the coefficients placed between two flags are the coefficients of a reference image according to the invention.

Preferably, the signal is in the form of consecutive packets, each packet header comprising a packet start field, the packet number in question, the type of wavelet transform used, a reference image field making it possible to indicate that the coefficients of the packet are reference image coefficients and are therefore not differences in coefficients between two images, a buffer number field indicating the number of the reference image having been used for coding the coefficients when a subsequence includes a plurality of reference images. Alternatively, the data signal further includes markers representative of the start and / or end of each determined video sub-sequence.

According to another alternative embodiment, the data signal further comprises for each other image, the number of the reference image on which each other image depends.

The data signal thus formed is similar to that previously described with reference to FIG. 2. It will not be further explained.

Fig. 6 shows the decoding algorithm according to the invention of a data signal representative of a sequence of video images. When the application is launched, the processor 100 of the device 10 reads from the read-only memory 102, the program instructions corresponding to steps E600 to E610 of FIG. 5 and loads them into random access memory 103 to execute them.

In step E600, the processor 100 of the device 10 determines in the data signal at least one field identifying coefficients associated with a reference image in the data signal.

It should be noted that the data signal to be processed comes from a peripheral device 107, such as a digital camcorder or any other means of acquiring or storing data.

The data signal to be processed can also come from a remote device via the communication network 113. The communication network 113 is for example an Internet type network or a telephone telecommunication network by which a videoconference type communication is established between two devices 10.

The communication network 113 can also be a Hertzian or satellite broadcasting network of coded video information according to the present invention. The determined identifier, the processor 100, at this same step, reads from the data signal, the coefficients identified by the marker.

This operation performed, the processor 100 determines in step E601 from the information contained in a predetermined field of the data signal the type of wavelet transform used to code certain areas of the image. Indeed, and according to a preferred mode, different types of wavelet transforms are applied to different parts of the image to be processed as a function of these.

For example, a wavelet transform known as Butterfly wavelets is assigned to the texture areas, a wavelet transform known as Loop wavelets is assigned to the texture areas. contours of natural objects, a wavelet transform known as Catmull-Clark wavelets is assigned to the contours of unnatural objects and finally a wavelet transform known as affine wavelets is assigned to singularities . The types of wavelet transforms applied during the determined coding, the processor 100 performs in step E602 the transforms into inverse second generation wavelets corresponding to the determined types for the corresponding coefficients. This step performed, the processor 100 reads in step E603 in the data signal to be processed and more particularly in the field of the mesh shape, the topology of the mesh of the image.

From the information obtained in steps E602 and E603, the processor 100 reconstructs the image in step E604 and transfers it for example to a display means such as the screen 104.

The processor 100 in step E605 determines the coefficients in the data signal of another image than a reference image. This image is an image contained in the same video sub-sequence as that comprising the reference image, the coefficients of which were read previously in step E600. For example, the processor 100 reads from the data signal the coefficients according to those read in step E600. These coefficients consecutive to the coefficients read in step E600 are then the coefficients of another image.

This operation carried out in step E606, the processor 100 makes the difference between the coefficients determined in step E600 and the coefficients determined in step E605.

This operation carried out, the processor 100 determines in step E607 from the information contained in the data signal the type of wavelet transform used to code certain areas of the image being processed. Indeed, and according to a preferred mode, different types of wavelet transforms are applied to different parts of the image to be processed as a function of these. However, it should be noted that this step is in many cases unnecessary. Indeed, during coding, the sequence of images has been divided into sub-sequences of images, the images of a sub-sequence of images are little different and thus the type or types of wavelet transforms used to code certain areas of the image being processed are identical to those determined in step E601 for the reference image of the same video sub-sequence. The processor 100 then performs in step E608 the transforms into inverse second generation wavelets corresponding to the types determined for the corresponding coefficients.

From the information obtained in steps E603 and E608, the processor 100 reconstructs the image in step E609 and transfers it for example to a display means such as the screen 104.

Thus, it should be noted that according to the invention, with only the mesh of an image, called the reference image, it is possible to reconstruct a set of images from it. This greatly reduces the amount of information needed to encode a video sequence and therefore the size of the data signal.

This operation performed, the processor 100 checks in step E610 if coefficients representative of other images are present in the data signal and associated with the reference image determined in step E600. If so, the processor 100 returns to step E605 and repeats the steps

E605 to E610 previously described as long as other coefficients representative of other images are present in the data signal.

If on the other hand, the processor 100 determines in the data signal a new marker identifying coefficients associated with a reference image in the data signal, the processor 100 returns to step E600 and performs the same operations with the coefficients of the new reference image than those described above.

It should be noted that the algorithm as described corresponds to a decoding algorithm conforming to the coding algorithm as described with reference to FIG. 2. When the data has been coded in accordance with the coding algorithm as described with reference to Figs. 4a and 4b, the reference images are no longer the images of the beginnings of sub-sequences and the other images are no longer following the reference images.

The processor 100 in addition to the steps of the previously described decoding algorithm must identify the reference images used for coding the other images. For example, the data signal is searched for markers representative of the start and / or end of video sub-sequences and the coefficients included between these markers are considered as the coefficients of the other video images of the sub-sequence. According to another alternative embodiment, the processor 100 searches in the data signal, for each other image, the number of the reference image on which each other image depends.

Of course, the present invention is not limited to the embodiments described here, but encompasses, quite the contrary, any variant within the reach of ordinary skill in the art.

Claims

1) Method for coding a sequence of video images, the sequence of video images being broken down into sub-sequences of video images, a mesh (E200, E203) being performed on the images of the video sequence, the mesh being composed of topology and geometry information, characterized in that the method comprises the steps of:

- association (E214, E521) of the topology information of an image of a sub-sequence of video images with all of the video images of the sub-sequence of video images,

transformations, for each image of the sub-sequence of video images, of geometry information of the mesh of the image of the sub-sequence of images into coefficients from at least one wavelet transform,

- Formation (E214, E521) of a data signal comprising the associated topology information and information representative of the coefficients.

2) Coding method according to claim 1, characterized in that the wavelet transform is a second generation wavelet transform.

3) Coding method according to claim 2, characterized in that the information representative of the transformed coefficients are the coefficients of at least one reference image of the sub-sequence and for each other image of the sub-sequence of video images , the information representative of the transformed coefficients is the difference between the coefficients of each other image and the coefficients of the reference image.

4) Coding method according to claim 2, characterized in that with a sub-sequence of video images is associated (E418) a predetermined number of reference images as a function of the number of images contained in the sub-sequence d 'video images.

5) Coding method according to claim 3, characterized in that the method further comprises a step of inserting information data into the signal to differentiate the coefficients of at least one reference image from the other coefficients. 6) Coding method according to any one of claims 1 to 5, characterized in that the mesh is a regular mesh.

7) Coding method according to any one of claims 1 to 6, characterized in that at least two different types of wavelet transforms are applied to at least two different regions of at least one image of the video sub-sequence .

8) Coding method according to any one of claims 2 to 7, characterized in that the video image sub-sequences in the video image sequence are determined by comparison (E207) of differences in image coefficients of the video sequence at a predetermined threshold.

9) Coding method according to any one of claims 2 to 7, characterized in that the video image sub-sequences in the video image sequence are determined by comparison (E408) of the signal-to-noise ratio between an image and a compensated image in motion with respect to a predetermined threshold.

10) Method for decoding a data signal representative of a sequence of video images, characterized in that the method comprises the steps of:

- determination (E600, E610) in the data signal, of images of a sub-sequence of video images,

- obtaining (E600, E605, E606) of the coefficients of the images of the sub-sequence of video images from information contained in the data signal, - transformation (E602, E608) of the coefficients obtained according to an inverse wavelet transform ,

- reconstruction (E604, E609) of each of the images of the video sub-sequence from the transformed coefficients and of a part of the mesh associated with the sub-sequence of video images and contained in the data signal.

11) A decoding method according to claim 10, characterized in that the part of the mesh associated with the sub-sequence of video images is the topology of the mesh of the sub-sequence of video images. 12) A decoding method according to claim 10 or 11, characterized in that coefficients, called coefficients of a reference image are obtained by reading the data signal and that coefficients called coefficients of other images are obtained by performing a difference between information contained in the data signal and the coefficients of the reference image.

13) decoding method according to any one of claims 10 to 12, characterized in that it further comprises a step of determining (E601), from information contained in the data signal, of at least one type of second generation inverse wavelet transform to be applied to at least part of the image coefficients of the video image sub-sequence.

14) Device for coding a sequence of video images, the sequence of video images being broken down into sub-sequences of video images, a mesh (E200, E203) being carried out on the images of the video sequence, characterized in what the device includes:

means of association (E214, E521) of topology information of an image of a sub-sequence of video images with all of the video images of the sub-sequence of video images,

means for transforming, for each image of the sub-sequence of video images, information on the geometry of the mesh of the image of the sub-sequence of images into coefficients from at least one wavelet transform,

- Means for forming (E214, E521) a data signal comprising the associated topology information and information representative of the coefficients.

15) Device for decoding a data signal representative of a sequence of video images, characterized in that the device comprises:

- means for determining in the data signal, images of a sub-sequence of video images, - means for transforming the coefficients of the images of the video sub-sequence obtained from the data signal according to a transform in reverse wavelets, means of reconstruction of each of the images of the video sub-sequence from the transformed coefficients and of a part of the mesh associated with the sub-sequence of video images and contained in the data signal.

16) Data signal representative of a sequence of video images, a mesh being made on the images of the sequence of video images, the mesh being composed of topology and geometry information, characterized in that the sequence d video images is divided into video image sub-sequences, the signal comprises for each video image sub-sequence information representative of the coefficients of the images of the video image sub-sequence and of topology information of the mesh of the video image sub-sequence.

17) Signal according to claim 16, characterized in that the information representative of the coefficients of the images of the sub-sequence of video images consist of the coefficients of at least one reference image of the sub-sequence of video images and for each other image of the video image sub-sequence of the difference between the coefficients of each other image and the coefficients of the reference image and in that the signal includes at least one predetermined piece of information identifying the coefficients of each image of reference of the sub-sequence of video images.

18) Data signal according to claim 16 or 17, characterized in that the coefficients of the images of the sub-sequence of video images are coefficients of a second generation wavelet transform of the mesh geometry of each of the images of the video image sub-sequence.

19) Computer program stored on an information medium, said program comprising instructions making it possible to implement the coding method according to any one of claims 1 to 9, when it is loaded and executed by a computer system .

20) Computer program stored on an information medium, said program comprising instructions making it possible to implement the method of decoding according to any one of claims 10 to 13, when it is loaded and executed by a computer system.