EP1299859A1

EP1299859A1 - Motion estimator for coding and decoding image sequences

Info

Publication number: EP1299859A1
Application number: EP01951762A
Authority: EP
Inventors: Guillaume Robert; Nathalie Laurent-Chatenet
Original assignee: France Telecom SA
Current assignee: Orange SA
Priority date: 2000-07-13
Filing date: 2001-07-03
Publication date: 2003-04-09
Also published as: FR2811791B1; US7502413B2; US20040047415A1; FR2811791A1; WO2002007099A1

Abstract

The invention concerns motion estimation and compensation between two given images, represented by a polygonal mesh, and provides a solution to mesh crowding and unmasking when the mesh is moved. More precisely, the invention concerns a method for coding a mesh representing an animated sequence image. Such a mesh is larger in size than said image.

Description

Motion estimator for coding and decoding image sequences.

The field of the invention is that of coding of images or sequences of images. More specifically, the invention relates to the estimation and compensation of the movement between two given images.

The concept of movement between two images of a sequence is a concept well known in the scientific and technical community of image processing. This is defined as the two-dimensional vector characterizing the difference in positions between a point of an image and the homologous point of the other image, on the assumption that these points correspond to the same physical point in each of the two images . The assignment of a displacement vector to each point of the image, or pixel, therefore defines a dense field of movement on the image, as illustrated in FIG. 1. The set 10 of the N * M vectors obtained for an image of size N * M is called the field of movement, or optical flow, between two images 11 and 12 corresponding respectively to the instants t and t + 1.

The estimation of the movement between two images (calculation of a displacement field or of disparities between two images) finds applications in all fields of imagery and image processing. We can cite, by way of examples, and without limitation: - video coding: the motion field is used for the prediction of an image from the previously decoded images. The representation by mesh of such a field is notably defined in different norms or standards, such as the MPEG-4 standard. medical imagery: analysis of the movement of the human body, heart, etc. object tracking (for example in the context of road traffic control).

The invention applies in particular to the processing of two-dimensional (2D) images, but also to the processing of images representative of multidimensional objects (in particular 3D). To date, there are several motion estimation techniques in image sequences.

A first technique consists in calculating a dense field of motion for the image, and in assigning a displacement vector to each pixel. A second technique aims to calculate a displacement vector per rectangular block of the image.

A third technique is based on the calculation of a mathematical model of the movement by region, possibly of arbitrary shape, of the image. Such a technique therefore also implements a segmentation of the image into regions, corresponding for example to the various video objects constituting the illustrated scene.

A final technique, described for example in patent application FR 2 783 123 entitled "Method for estimating the movement between two images", consists in calculating a field defined by a finite element model. A mesh is associated with the image, and an interpolation or extrapolation function makes it possible to calculate the value of the field at each point of the image, inside each of the meshes, as illustrated by the figure. 2.

This technique is also based on the implementation of a differential method, which determines the movement parameters by optimization of a mathematical quality criterion (for example, a quadratic error between the image and its value predicted by the compensation of the movement). Such an optimization is carried out using a differential mathematical optimization method, in the case of criteria having the necessary mathematical properties (for example, a differentiable criterion). Thus, for example, a triangular mesh is associated with the image 21. The value of the motion vectors of the vertices referenced 22, 23 and 24 of the mesh is optimized. Then one carries out an interpolation (or an extrapolation) in order to determine the vectors of movement of the points not corresponding to vertices of the mesh, such as the points referenced 25 and 26 for example. Such a motion estimation method, based on the implementation of a hierarchical mesh representative of the image to be coded, is particularly suitable for the progressive transmission of data, and therefore represents an advantageous solution to the problems of band saturation. bandwidth and adaptability to communication networks, of different natures and capacities. However, such a method encounters certain difficulties when tracking movement.

Thus, a drawback of this technique of the prior art is that it generates displacements of meshes, causing settlement or uncovering on certain areas of the image, as illustrated in FIG. 5. In fact, the deformable meshes define a continuous representation of a movement field (that is to say that the mesh follows the moving objects in the scene), while the real movement of a video sequence is of a discontinuous nature. Consequently, when an object moves for example from left to right in the image 51, the left zone 52, discovered by the displacement of the object, represents new information, and, therefore, is not more meshed. The right-hand area 53 supports a compaction of meshes. Similarly, when different planes and objects overlap in a scene, occultation zones appear, generating lines of discontinuity. However, in addition to the movement information, the vertices of the mesh also carry photometric and / or colorimetric information which make it possible to generate a photometric and / or colorimetric field, and therefore to approximate the image. The appearance of non-meshed discovery areas therefore corresponds to the entry into the image of areas whose photometry and / or colorimetry cannot be approximated, that is to say black areas.

The solutions generally proposed to compensate for the appearance of non-meshed discovery areas, or settlement areas, characterized by a redundancy of information, consist in forcing the nodes of the edge of the image (for example the node referenced 511) to stay still. A disadvantage of this technique of the prior art is that the approximation of the incoming zones is of poor quality, due to the stretching of the associated meshes.

Another disadvantage of this technique of the prior art is that the motion estimation is biased by the immobility constraint imposed on the nodes of the edges of the image.

This technique of the prior art also has the drawback of causing an overly approximation of the areas leaving the image that is too costly, corresponding to the areas of compacting of the meshes. Indeed, in the settlement zones, one obtains an over-representation of the field of movement, because too many meshs are used to make the approximation of a reduced portion of the field of movement. Such an over-representation does not harm the quality of approximation of the field of motion, but generates an additional cost of transmission. Another solution envisaged to compensate for the appearance of the areas of uncovering or of areas of settlement of the mesh consists in inserting new points on the black areas and in constructing a constrained Delaunay mesh.

A disadvantage of this technique of the prior art is that it is costly in terms of time and speed of transmission and / or storage of information.

Another problem linked to disturbances in the field of motion is the appearance of inversion of the meshes of the hierarchical mesh representative of the image, as illustrated in FIG. 3, when certain meshes 31, 32 move one relative to the 'other according to antagonistic motion vectors 33, 34. Several methods have been considered to overcome such reversals of meshes.

In particular, consideration has been given to placing constraints on the various meshes, so as to prohibit the phenomena of reversal.

One also envisaged to carry out a postprocessing on the mesh, which can be carried out according to two distinct modes. A first embodiment of a postprocessing consists in applying the motion vectors as estimated to the various nodes of the mesh, then in detecting the motion vectors having caused the appearance of defects in the mesh, and finally in correcting their value, so as to compensate for the mesh reversal phenomena. A second embodiment of a postprocessing consists of an iterative process: with each iteration, one applies a part of the estimated displacement to each of the nodes of the mesh, so as not to generate a reversal of meshes. The iterations are then repeated until a convergence of the process is obtained. However, the methods of postprocessing acting after the vectors of movement of the various nodes of the mesh were estimated, they do not allow an optimal management of the reversals of meshes. Indeed, in the post-processing methods, the motion vectors are corrected independently of their contribution to the minimization of the prediction error (for example of the quadratic error between the image and its value predicted by motion compensation ).

The technique described in French patent application FR 99 15568 entitled "Method for estimating movement between two images with management of mesh reversals and corresponding decoding method" offers a solution to the problem of reversal generated by the restrained motion estimator on the implementation of a hierarchical network.

Such a technique consists in undertaking an initial optimization of the motion vectors of the mesh, by letting the motion estimator create possible reversals between two successive instants t _λ and tj, so as to detect the discontinuity zones thus generated. The method then consists in carrying out a new estimation of the movement between the instants t _x and t ₂ , by excluding the faulty areas (that is to say the areas containing at least one mesh reversal), in order to minimize the prediction error between the two images corresponding to the instants t _j and t ^ considered. This new estimation makes it possible to determine the optimal motion vectors for the continuous area of the image (that is to say admitting a bijection between t _j and ^) and thus to avoid the values of the motion vectors obtained at during the initial optimization are not disturbed by the existence of discontinuity zones. The faulty areas are then approximated by a frequency or spatial method, when the method is applied to image compression for example, and permanently excluded from the optimization method, when the technique is applied to the monitoring of video objects for example .

A drawback of this technique of the prior art is that it does not make it possible to manage the appearance of areas of discoveries and areas of compacting of meshes during a translation of an object within the image, as described above.

The invention particularly aims to overcome these drawbacks of the prior art. More specifically, an objective of the invention is to provide a technique for coding images represented using a mesh, allowing estimation and compensation of the movement within the image.

Another objective of the invention is to implement an image coding technique making it possible to obtain a good approximation of the photometric and / or colorimetric surface of the image, in particular for highly textured areas.

The invention also aims to provide a simple and robust image coding technique.

The invention also aims to implement an image coding technique suitable for all areas in which the movement is estimated using meshes, and in particular the field of video coding (such as the MPEG4 and H263 + for example).

Another object of the invention is to provide a technique for coding images presenting a reduced cost of transmitting information. Yet another objective of the invention is to implement an image coding technique ensuring visual fluidity of the movement of the image.

Another object of the invention is to provide an image coding technique making it possible to effectively manage the appearance of areas of discovery and or of settlement zones during the movement of the objects constituting the image.

Another object of the invention is to implement an image coding technique making it possible to manage the phenomena of inversion and mesh concealment. Another object of the invention is to provide an image coding technique making it possible to reduce the transmission and / or storage of motion vectors for the areas of mesh settlement.

Yet another objective of the invention is to implement an image coding technique making it possible to ensure the consistency of the ratio of the information transmission rate to the image distortion rate.

These objectives, as well as others which will appear subsequently, are achieved, according to the invention, using a method of coding a mesh representative of an image of an animated sequence, such a mesh being larger than said image. By image is meant here a two-dimensional image, or a multidimensional image, for example representative of a 3D object.

Thus, the invention is based on a completely new and inventive approach to the estimation and compensation of movement for the coding and decoding of image sequences, allowing the management of discoveries and settlements of meshes during displacement. objects. Indeed, the invention in particular proposes the coding of a mesh of dimension greater than that of the image with which it is associated, and therefore goes against the prejudices of the skilled person, who has always sought to reduce the amount of information to be coded and transmitted and / or stored on a data carrier. It seems indeed completely useless, a priori, for those skilled in the art of constructing a mesh larger than the image, since no information is available beyond the limits of the image.

The invention thus makes it possible, during motion tracking, to maintain a mesh over all of the images in a video sequence. Advantageously, said image having an area of N * M pixels, said mesh has an area of at least nN * nM pixels, n being an integer greater than or equal to 2.

The invention ideally provides for using an infinite size mesh. By default, the mesh extends on either side of the image, so that its surface is at least n times greater than that of the image, where n is a sufficiently large integer, defined before the implementation of the coding method according to the invention.

According to a first advantageous characteristic of the invention, said mesh is a hierarchical mesh having at least two levels of mesh nested within a pyramid of mesh.

Thus, the use of a nested hierarchical mesh 41 makes it possible to obtain a robust coding, and reduces the cost of transmission and / or storage of information. Indeed, there is a strong correlation between the information carried by a parent node and that carried by its child nodes within the mesh. Therefore, the values of the nodes 42, 43 belonging to the highest level of the mesh pyramid (corresponding to a coarse mesh of the image) are quantified, with a view to their transmission, for example to an encoder, and / or their storage on a data carrier 45. On the other hand, for the lower levels of the pyramid of mesh (corresponding to a finer mesh of the image), only the differential values of the child nodes 44 are quantified, in view their transmission and / or storage, as illustrated in FIG. 4.

According to a second advantageous characteristic of the invention, said mesh is a triangular mesh consisting of an arrangement of vertices and triangular faces, each defined by three references to the vertices which it connects, and having three edges each connecting two of said vertices. The invention is of course also applicable to any other type of polygonal mesh, which may have undergone prior triangulation.

According to an advantageous technique of the invention, such a coding method associates with at least some of the vertices and / or some of the faces of said mesh at least one of the following markers: an "inside" marker, a face carrying an "inside" marker if it has a non-empty intersection with said image, and a vertex bearing an "interior" marker if it belongs to an "interior" face; an "incoming" marker, a vertex carrying an "incoming" marker if it has an "inside" marker and is outside the image, and a face carrying an "incoming" marker if at least one of the three vertices that it links carries an "incoming" marker.

The introduction of such markers makes it possible to differentiate the vertices of the mesh, and therefore makes it possible, during the motion estimation, to apply a processing adapted to each of the vertices and / or of the faces of the mesh, according to the marker which it is associated. Advantageously, such a coding method implements the succession of following steps: estimation of the movement of said vertices of said mesh, between two successive images of said sequence; construction of colorimetric and / or photometric information intended for vertices and / or faces of said mesh carrying said marker

"Incoming".

Such a construction step can be implemented according to the approach proposed in patent document FR-98 12525 having for title "coding process of still or moving images with reduction and adaptation of the bit rate". This technique relates to a method of coding an image digital, aimed at producing a binary train representative of this image, the length of the binary train being a function of desired representation. This method comprises the following steps: defining, on a domain of the image to be coded, a hierarchical mesh comprising a plurality of nested meshes whose mesh vertices can be pixels of said image; perform optimizations of luminance, chrominance, and positions on each mesh level; determine, for each mesh of said hierarchical mesh, a luminance difference between the image to be coded and an interpolated image obtained from the vertices of the nested mesh to which the mesh in question belongs, and introduce the values (advantageously differential coded) into the binary train compared to the preceding hierarchical level) of positions, luminance and chrominance of the vertices of the meshes whose luminance difference is greater than a threshold difference.

According to this technique, at the end of the meshing step, a quaternary tree structure, associated with the hierarchical mesh, is constructed to manipulate the values (colors and positions) of the vertices of the meshes. The tree presents a number of nodes equal to the number of triangles in the corresponding level of mesh. Each node of the tree relates to a single triangle of the hierarchical mesh.

Once this tree has been constructed, the data of the tree to be introduced into the bit stream representative of the image which will be transmitted and / or stored is selected. This selection depends on the desired quality.

To make this selection, a luminance difference between the image to be coded and the interpolated image is calculated for each triangle from the vertices of the nested mesh to which the mesh in question belongs. This difference is then compared to a threshold difference for each triangle. The value of the threshold deviation depends on the quality of representation desired. The part of the tree relating to the triangles whose luminance difference is greater than the threshold difference is then introduced into the binary train, which therefore makes it possible to transmit the data relating to the image as a function of the local quality of these different triangular partitions. Preferably, said estimation step comprises the following sub-steps: optimization of motion vectors of the vertices of said mesh carrying said "interior" marker, so as to minimize a predetermined error criterion for reconstruction of the following image; - interpolation and / or extrapolation of motion vectors from the vertices of said mesh not carrying said "interior" marker, in order to fluidize the overall displacement of said mesh.

Thus, during the motion estimation, only the meshes included at least partially in the image (that is to say the faces of the mesh carrying the "interior" marker) are used to determine the motion vectors. The external nodes of the mesh (that is to say the nodes not carrying the marker "interior") perceive meanwhile motion vectors interpolated and / or extrapolated. In this way, when the motion vectors are applied to the mesh, the settlement effects disappear and the areas discovered are covered by external meshes.

According to a preferred embodiment of the invention, said interpolation and / or extrapolation step implements a double iterative front-back scanning, applied to the vertices of the lowest level of said grid pyramid for which said vectors movement of said vertices have been optimized.

Indeed, only the vertices of the mesh carrying an "interior" marker have a significant motion vector. The vertices of the mesh that do not belong to the image have an indefinite motion vector. A problem therefore arises when using these motion vectors to implement a global displacement of the pyramid of mesh. In fact, assigning them a zero motion vector risks generating settlement of meshes at the edge of the image, and biases the fluidity of displacement of the image, obtained by global optimization of the motion vectors.

The invention therefore provides for artificially propagating the motion vectors from the vertices carrying an "interior" marker to the vertices not carrying an "interior" marker, by implementing such an iterative front-rear scan.

Advantageously, after said estimation step, such a coding method implements a step of moving each of the vertices of said mesh, by applying to it its own motion vector.

According to an advantageous characteristic of the invention, said construction step implements on the one hand a finite element coding model, and on the other hand an alternative coding model, the latter being implemented for at least certain zones. of said image, according to a predetermined error criterion.

Indeed, when implementing a hierarchical coding, for example as described in the French patent application FR 98 12525, and when a portion of the image is very textured, it is then necessary to provide a number of large mesh levels to obtain photometric and / or colorimetric information of sufficient quality. In this case, the efficiency of hierarchical coding is low. In other words, hierarchical coding is well suited for relatively simple images, but not for images with highly textured parts.

We therefore envisaged a new approach, described for example in French patent application FR 99 06815 entitled "Hierarchical coding method and selective implementation of a coding based on reversible transformation, and corresponding decoding method". Such a method of coding an image to be coded comprises the following steps: definition of a hierarchical mesh having at least two nested mesh levels formed of meshes defined by vertices (which may be pixels of said image to be encoded); determination, for each of said meshes, of error information between said image to be coded and an interpolated image obtained from the vertices of the meshes belonging to the mesh level of the mesh considered; stopping the refinement of the meshes having error information lower than a first predetermined threshold; implementation of a specific coding for the meshes presenting an error information higher than a second predetermined threshold; - Continuation of the refinement of the meshes having error information greater than said first predetermined threshold and less than said second predetermined threshold.

Thus, according to this method, two distinct coding modes are selectively used. The hierarchical coding, or with nested meshes, is the main coding, or basic coding, but it is not used systematically: for the image portions which require it, another more efficient coding is used.

The decision to use this specific coding is made by comparison with a threshold. More precisely, at each node considered in the tree representative of the hierarchical mesh, three possibilities are offered: stop the refinement, or the division, of the meshes (the quality reached for the corresponding image portion being sufficient), pass to the hierarchical level next, keeping the same hierarchical coding, or use another coding (in particular a coding better suited to highly textured portions).

Note that it is possible to use several different types of specific coding (or the same coding with different parameters), to adapt the choice of coding even more precisely to the characteristics of the image portion.

According to a preferred embodiment of the invention, said construction step implementing a hierarchical coding model, it comprises the following sub-steps: temporary normal projection of said vertices carrying said marker

"entering" on an edge of said image; optimization of said colorimetric and / or photometric information of said vertices and / or of said faces of said image, so as to minimize a predetermined error criterion for reconstruction of the following image; repositioning of said vertices carrying said "incoming" marker to their position before projection, said "incoming" vertices carrying said optimized colorimetric and / or photometric information.

Such a temporary normal projection of the vertices carrying the “incoming” marker on the edges of the image makes it possible to avoid taking account of pixels not belonging to the image, and therefore ensures the stability of the step of optimization of colorimetric and / or photometric information.

Preferably, said specific coding comprises the following sub-steps: - construction of a symmetrized image, obtained by applying to said image an axial symmetry with respect to each of its edges and a central symmetry with respect to each of its corners; optimization of said colorimetric and / or photometric information of said vertices and / or of said faces of said symmetrized image by application of said alternative model (DCT, fractals, matching pursuit).

The normal projection of the "incoming" vertices of the mesh on the edges of the image not being suitable for the approximation of high frequencies such as that implemented by application of the specific coding, it was envisaged to carry out a symmetrization of the image. The optimization processing is then carried out on the symmetrized image, and a solution is thus provided to the problem of taking into account pixels not belonging to the image.

According to an advantageous technique of the invention, such a coding method implements a step of memorizing photometric and / or colorimetric information associated with the vertices and / or the faces emerging from said image. Thus, the information relating to the zones leaving the image can be reused if such zones are brought to return to the image, without it being necessary to apply a new treatment to them. Such a situation is frequent in a video sequence: it is for example the case of a decoration located in the background of a character followed by a camera, if the character goes back and forth in the image.

According to a first advantageous variant of the invention, as long as one face of said mesh carries the "incoming" marker, said construction step implements at least one iteration of the following successive sub-steps: - optimization of said colorimetric information and / or photometrics of said vertices of said "incoming" face, so as to minimize a predetermined error criterion; transmission to a terminal and / or storage on a data medium of said optimized colorimetric and / or photometric information. Thus, the nodal photometric and / or colorimetric values of the incoming meshes are re-optimized, then transmitted and / or stored as long as they are not fully entered in the image.

According to a second advantageous variant of the invention, as long as one face of said mesh carries the "incoming" marker, said construction step implements at least one iteration of the following successive sub-steps: elaboration of a criterion making it possible to evaluating an image reconstruction quality obtained by taking into account the current photometric and / or colorimetric information of the vertices of said face; when said quality is not satisfactory: - optimization of said colorimetric and / or photometric information of said vertices of said face, so as to minimize a predetermined error criterion for reconstruction of the following image; transmission to a terminal and / or storage on a data medium of said optimized colorimetric and / or photometric information; transmission and / or storage of a tree of markers indicating to said terminal whether said colorimetric and / or photometric information of said vertices of said "incoming" face have been optimized or not. The quality of the surface approximation obtained is thus estimated taking into account the available photometric and / or colorimetric nodal values, and optimization and transmission and / or storage of the optimized values are only practiced in the event of a significant error. . Such a variant makes it possible to reduce the amount of information to be transmitted and / or stored. The implementation of such an alternative embodiment requires the transmission and / or storage of a tree of markers, indicating, for example to the decoder, whether the photometry and / or the colorimetry of the incoming mesh is to be updated before rendering the image. According to a first advantageous embodiment of the invention, said step of estimating the movement further comprises a sub-step of constructing a tree of the motion vectors to be transmitted and / or stored, the belonging of a motion vector to said tree being determined from a criterion of relevance of the movement applied to said faces of said mesh. Indeed, in the settlement zones, an over-representation of the field of motion is obtained. It is therefore particularly advantageous to transmit only some of the motion vectors associated with such settlement zones, so as to reduce the cost of transmission and / or storage of information.

Preferably, said relevance criterion results from a predetermined weighted combination of the following sub-criteria: a surface sub-criterion, making it possible to evaluate the ratio between the surface of a face and the average surface of the faces of said mesh belonging to the same level of said mesh pyramid; a compactness sub-criterion, making it possible to evaluate the relationship between the surface and the perimeter of a face of said mesh; a sub-criterion of antagonism of the movements, making it possible to evaluate the antagonism of the motion vectors of vertices of said mesh.

According to a second advantageous embodiment of the invention, all the motion vectors of said vertices of said mesh are transmitted and / or stored, a suitable decoder identifying according to a predetermined criterion the vectors to be taken into account.

The cost of coding the image is thus reduced, since it is at the time of decoding that the relevance of the motion vectors transmitted and / or stored is evaluated. According to a first advantageous technique of the invention, in the presence of a concealment of at least one apex obscured by a face comprising at least one obscuring apex, one proceeds to a fusion of said obscured apex with said obscuring apex.

The vertex resulting from the fusion of the obscured vertex and the obscuring vertex is placed on the edge initially connecting the obscured vertex and the occulting vertex, for example in its middle.

According to a second advantageous technique of the invention, such a coding method implements a step of detecting at least one area of said image obscured during movement of said mesh, and a step of storing photometric information and / or colorimetric relating to said obscured area, in view of possible future use, said information not being processed.

The invention also relates to a decoding method and to coding and decoding devices of a mesh representative of an image of an animated sequence, said mesh being of size larger than said image.

The invention also relates to a signal representative of a mesh representative of an image of an animated sequence, said mesh being of size larger than said image.

Other characteristics and advantages of the invention will appear more clearly on reading the following description of an embodiment preferential, given by way of simple illustrative and nonlimiting example, and of the appended drawings, among which: FIG. 1, already described above, illustrates the concept of field of motion; - Figure 2, commented above, describes an example of approximation of the field of motion of Figure 1 by interpolation or extrapolation of inter-nodal movements; FIG. 3, described above, presents an example of mesh reversal, resulting from the existence of antagonistic motion vectors within the motion field presented in FIG. 1; FIG. 4, described above, illustrates an exemplary embodiment of hierarchical transmission of information; FIG. 5, described above, shows an example of the appearance of a zone of discovery and a zone of compaction of the mesh resulting from a movement within the image; FIG. 6 illustrates the construction of a mesh of dimension greater than that of the image; FIG. 7 presents an example of “interior” markers associated with certain faces and certain vertices of a mesh as illustrated in FIG. 6; FIG. 8 illustrates the implementation of a temporary normal projection of vertices during the stage of construction of colorimetric and / or photometric information; FIG. 9 shows an example of image symmetrization for the implementation of an alternative coding; FIG. 10 illustrates an example of propagation of the motion vectors from the vertices carrying the marker "interior" towards the other vertices of the mesh, with a view to a global displacement of all the vertices of the pyramid of mesh; - Figure 11 presents an example of meshes carrying the marker "incoming" during several frames of information; FIG. 12 illustrates a phenomenon of packing of meshes on a zone of occultation of the mesh; FIG. 13 presents a first example of management of occultation phenomena by fusion of vertices; FIG. 14 illustrates a second example of managing the occultation phenomena by implementing a 3-manifold mesh. 1. Basic principles of the invention

The invention therefore relates to an improvement in the motion estimation technique implementing a hierarchy of nested meshes (or hierarchical coding), as described in the preamble. According to the present invention, provision is in fact made for using a mesh of size greater than the images of the sequence.

We present, in relation to FIG. 6, an embodiment of a mesh of dimensions greater than those of the image.

An image 61 is described using a triangular mesh, consisting of a plurality of faces and vertices, which extends beyond the limits of image 61. Thus, some vertices 611 are located at the 'inside the image 61, and other vertices 612 are placed outside the limits of the image 61. Ideally, such a mesh is a hierarchical mesh nested with infinite surface.

Furthermore, the concept of markers, associated with the faces and / or the vertices of the mesh, is introduced, with a view to reducing the data transmission rate necessary for coding. Figure 7 shows an example of marking the vertices and the faces of a mesh using an "interior" marker.

Consider image 71 represented using a mesh 72. Each vertex and / or face of the hierarchy of mesh 72 is, or is not, marked by means of an "interior" marker (indicated by the two letters IN in FIG. 7) depending on whether it is located inside or outside of image 71. More precisely, a face of the mesh is "inside" if it has a non-empty intersection (in number of pixels) with the image 71. Thus, the face referenced 721 bears the indication IN in FIG. 7.

A vertex of the mesh 72 is "inside" at a given level of the mesh 72 if it belongs to an "inside" face of the level considered. Thus, the vertex referenced 722 bears the indication IN because it belongs to the "interior" face referenced 721.

It is therefore possible that a vertex is "interior" (that is to say bears the indication IN in FIG. 7) while being outside of image 71. For example, the vertex referenced 723 is " interior "because it belongs to the" interior "face referenced 721, but it is exterior to image 71.

We also define an "incoming" marker, which can be associated with a face and / or a vertex of the hierarchical mesh 72. We consider that a vertex is "incoming" if it is "inside" and that its photometric information and / or colorimetric are undefined or unsatisfactory, that is to say if it is "inside" but it is located outside of image 71. We also consider that one face of the mesh 72 is "incoming "if at least one of its vertices is" incoming ". 2. Management of mesh stripping zones

2.1. Algorithmic description At each estimation of the movement of the image considered, all the vertices of the mesh carrying the "interior" marker are used to determine the optimal motion vectors of the mesh. The motion vectors of the neighboring vertices, not carrying the "interior" marker, are then interpolated or extrapolated, in order to make the displacement of the mesh more fluid. It can of course be envisaged that certain motion vectors are interpolated, and that other motion vectors are extrapolated.

In addition, the photometric and / or colorimetric information associated with the faces and / or the vertices of the mesh entering the image for the first time (that is to say bearing the marker "entering") is updated, in order to of make a complete approximation of the photometric and / or colorimetric surface of the image.

2.2. Photometric and / or colorimetric approximation of the image surface by finite elements Such a step of constructing photometric and / or colorimetric information of the faces and / or vertices "entering" the mesh implements an optimization by finite element model, according to the coding method described in patent application FR 99 06815 entitled "Hierarchical coding method and selective implementation of a coding based on reversible transformation and corresponding decoding method".

Such an optimization notably consists in carrying out optimizations of luminance, chrominance and positions on each level of mesh. Then, for each mesh of the hierarchical mesh, a luminance difference between the image to be coded and an interpolated image is obtained, obtained from the vertices of the nested mesh to which the mesh considered belongs. Finally, we introduce into a binary train representative of the image the values (advantageously coded in differential compared to the previous hierarchical level) of positions, luminance and chrominance of the vertices of the meshes whose luminance difference is greater than a difference threshold. In order to avoid taking into account pixels not belonging to the image, and therefore to achieve better optimization of the colorimetric and / or photometric information, the procedure is, according to the invention, a temporary normal projection of the vertices " entering "on the edges of the image, as illustrated in FIG. 8. Thus, the vertices 821a, 822a and 823a of the representative mesh of the image 81 are projected onto the edges of the image at positions 821b, 822b, 823b.

When certain faces of the mesh, after projection, do not contain any more pixels (i.e. one obtains empty faces after projection), they are not taken into account for the optimization of the information of colorimetry and / or photometry. The vertices of these empty faces are however taken into account at during the optimization of colorimetric and / or photometric information associated with neighboring aces of the mesh.

After such an optimization, the projected vertices 821b, 822b, 823b are returned to their initial position, respectively 821a, 822a and 823a, but they are associated with the optimized color and / or photometric information after projection.

2.3. Photometric and / or colorimetric approximation of the textured areas of the image

2.3.1. Implementation of a specific coding The finite element approximation as described above is well suited to smooth images. On the other hand, when portions of the image are very textured, it is necessary to implement a high number of levels of the hierarchical mesh, to obtain a satisfactory surface approximation. The hierarchical coding efficiency then becomes low. To overcome this drawback, it has been envisaged, in French patent application No. FR 99 06815 entitled "Hierarchical coding method and selective implementation of a coding based on reversible transformation and corresponding decoding method", to resort to an alternative model on the residual part of the textured zones.

The technique described in this patent application, and already recalled above, relates to a method of coding an image to be coded, comprising the following steps: definition of a hierarchical mesh having at least two nested mesh levels formed of meshes defined by vertices (which may be pixels of said image to be coded); - Determination, for each of said meshes, of error information between said image to be coded and an interpolated image obtained from the vertices of the meshes belonging to the mesh level of the mesh considered; stopping the refinement of the meshes having error information lower than a first predetermined threshold; implementation of a specific coding for the meshes presenting an error information higher than a second predetermined threshold; continuation of the refinement of the meshes having error information greater than said first predetermined threshold and less than said second predetermined threshold.

Thus, according to such a technique, two distinct coding modes are selectively used. The hierarchical coding, or with nested meshes, is the main coding, or basic coding, but it is not used systematically: for the image portions which require it, another more efficient coding is used (DCT, fractals, matching pursuit, etc.).

We therefore choose, according to the invention, to adopt an approach similar to that described in French patent application FR 99 06815, and to implement specific coding for the highly textured areas of the image. 2.3.2. Image symmetry

The implementation of a projection of the "incoming" vertices of the mesh, prior to the optimization of the photometric and / or colorimetric information, is not suitable for the use of such specific coding. Indeed, such an operation of projection of the vertices strongly distorts the approximation of the high frequencies of the image.

We therefore thought of operating a symmetrization of the image, as illustrated in FIG. 9. The images referenced 93, 95, 97 and 99 are thus obtained by applying to image 91 an axial symmetry with respect to each of the edges. of the image 91. The images referenced 92, 94, 96 and 98 are images of the image 91 by central symmetry with respect to each of the vertices of the image 91.

One then proceeds to an optimization of the photometric and / or colorimetric information from the image 90, obtained by regrouping of the images referenced 91 to 99. One thus solves the problem of the pixels of the faces

"incoming" of the mesh, outside the image. One thus applies a treatment similar to all the “entering” faces of the mesh. 2.4. Motion estimation

2.4.1. Estimation of motion of the "interior" vertices We proceed, for the estimation of motion, according to a technique similar to that described in paragraph 2.2 for the surface approximation by finite elements. The motion estimation is therefore carried out on the faces and the vertices of the mesh carrying the "interior" marker. The "incoming" vertices are temporarily projected normally on the edges of the image, one then proceeds to an optimization of the motion vectors of the vertices of the mesh, then, the "incoming" vertices projected return to their initial position before projection. 2.4.2. Propagation of movement

Only the vertices of the mesh carrying the "interior" marker have a significant motion vector. It is thus necessary to determine the value to be allotted to the vectors of movement of the vertices not carrying the marker “interior”, in order to be able to use them during the total displacement of the pyramid of mesh. Assigning null motion vectors to vertices that are not "interior" indeed causes the appearance of mesh settlement zones, and alters the fluidity of the displacement determined by global optimization of motion vectors.

We therefore considered adopting an approach inspired by that of the chamfer distance map calculations, as described in the thesis document of the Joseph Fourier University of Grenoble, defended on September 21, 1994 by Edouard Thiel and entitled "Les chamfer distances in image analysis: foundations and applications ". Such an approach consists in propagating the motion vectors from the "inside" vertices towards the non "inside" vertices, by applying a double iterative front-back scanning to the vertices of the lowest level of the mesh pyramid, noted L _m , where the motion vectors have been optimized. This avoids stretching of meshes, by allowing the insertion of incoming zones in the image, and compaction of meshes, avoiding the exclusion of outgoing zones.

Such propagation of the movement is, for example, implemented in a particular embodiment, according to the following algorithm: For all vertices S of L _m if INTERIOR (S) then S becomes PROPAGATED otherwise S becomes non-PROPAGATED Iterate as long as there are undefined motion vectors For all vertices S of L _m traveled from top-left to bottom- right

If non-PROPAGATED (S) and S has at least 2 neighbors PROPAGATED Then MOTION (S) = average of the MOVEMENTS of neighbors PROPAGATED S becomes PROPAGATED For all the vertices P of the traversed pyramid of Lm. _! at L ₀ MOVEMENT (P) = MOTION (F) such that P is father of F

An example of the implementation of such an algorithm is presented in relation to FIG. 10. We consider the mesh 101 representative of image 102. All the "inside" vertices of image 102 are assigned an optimized motion vector, represented by an arrow. We then proceed to a scan 103 of the image aimed at propagating the movement from top to bottom of the image 102. The vertices 1010 and 1011 of the mesh 101 are "inside": their respective motion vectors are therefore optimized. According to the algorithm presented above, the vertex referenced 1012 of the mesh 101 having the vertices referenced 1010 and 1011 for neighbors, a motion vector is assigned to it, obtained by averaging the motion vectors of the vertices referenced 1010 and 1011.

Next, a scan 104 of the image 102 is carried out with the aim of propagating the movement from the bottom to the top of the image. We thus attribute to the referenced vertex

1014 a motion vector obtained by averaging the motion vectors of its neighboring vertices, referenced 1012 and 1013, whose motion vectors had been determined during the first scan 103.

We then apply a new scan referenced 105 aimed at propagating the movement from the top to the bottom of the image 102. The movement is propagated for example at the vertex referenced 1016 of the mesh, to which we assign a vector of average motion from the motion vectors of the vertices referenced 1014 and 1015.

Finally, we carry out a scan referenced 106 from bottom to top of the image 102. At the end of such a scan, we see that all the vertices of the mesh 101 have a motion vector, represented using an arrow in FIG. 10. The movement has therefore been propagated to the whole of the mesh 101. The parent nodes of the nodes of the mesh 101, in the upper levels of the mesh pyramid, are then associated with motion vectors of values equal to those of their child nodes. 2.4.3. Overall displacement of the mesh pyramid

We then proceed to a global displacement of each vertex of the mesh pyramid, by applying to it the motion vector which was associated with it during the previous step of motion propagation. The effect of such a displacement is to: - introduce meshes when an object enters the image; make mesh out by following an object emerging from the image; limit the degeneration of the mesh.

2.5. Photometric and / or colorimetric approximation of the faces carrying the “incoming” marker on several successive frames We note, in the field of video coding, that one face of the mesh, which for example follows the entry of an object into the image , can carry the "incoming" marker on several successive data frames, as illustrated in FIG. 11.

However, the "incoming" vertices of the mesh do not have photometric and / or colorimetric information, the latter being approximated from the photometric and / or colorimetric information available, associated with the portions of the "incoming" faces located inside the image.

According to the invention, two alternative embodiments are proposed for managing the photometric and / or colorimetric information of the faces carrying the "incoming" marker on several successive frames. A first alternative embodiment consists in optimizing this information at each data frame. Thus, the nodal photometric and / or colorimetric values of the incoming faces 110 are reoptimized and transmitted and / or stored on a data medium each time the photometric and / or colorimetric surface of the image is approximated (i.e. at each instant t, t + 1, t + 2 and t + 3), as long as the incoming face 110 considered is not entirely included in the image 111.

A second alternative embodiment, less costly in terms of data transmission, consists in establishing a quality criterion, and in comparing, with this criterion, the quality of the photometric and / or colorimetric approximation of the surface of the image, obtained in taking into account the current nodal photometric values of the incoming faces. If such a comparison reveals a significant error in the approximation of the image, one then proceeds to an optimization and a transmission and / or storage on a data carrier of the photometric and / or colorimetric information of the "incoming" vertices. If, on the other hand, the quality of the approximation of the image is satisfactory, the current photometric and / or colorimetric values are kept for the "incoming" vertices.

According to this second variant, a tree of specific markers is also constructed, associated with the "incoming" vertices 112 to 114, making it possible to indicate to a decoder or to any other suitable terminal, whether the photometry and / or the colorimetry of the face " incoming "considered 110 must be updated before rendering image 111.

2.6. Memorization of photometric and / or colorimetric information associated with the zones leaving the image At the end of the preceding steps of motion estimation and approximation of the photometric and / or colorimetric surface of the image, a memorization is carried out. photometric and / or colorimetric information associated with the outgoing areas of the image, so that such information is directly available, without the need to implement a new long and costly treatment, in case such outgoing areas re-enter the image.

Indeed, it is frequent in video sequences that certain objects enter and leave the image on several occasions (for example a decoration located in the background of a character performing back and forth in the image), and it is therefore particularly clever to memorize information relating to this object, so as not to have to implement a new processing each time the object enters the image.

3. Management of mesh settlement zones 3.1. Principle of adaptive motion optimization

According to the principle of hierarchical coding by finite elements described above, the motion vectors of the vertices belonging to the levels L ₀ to L _m of the mesh pyramid are all optimized.

However, when two objects 121, 122 intersect within the image 120 for example, it appears zones of compaction of meshes 123 at the border between the two objects, as illustrated in FIG. 12. In such zones of settlement 123, an unnecessary and costly over-representation is obtained in terms of data transmission of the motion field.

To compensate for this phenomenon, it has been envisaged, according to the invention, to introduce a criterion of relevance of the movement, making it possible to select the most relevant motion vectors, and which make it possible to best estimate the displacement within the zone. of settlement.

We therefore define a "relevant" marker, carried by the vertices of the mesh whose motion vectors must be transmitted and / or stored on a data medium. We then construct a tree grouping together all the motion vectors to be transmitted and / or stored, which allows, in the settlement zones, to effect motion transmission on a locally reduced number of levels of the mesh pyramid.

According to another alternative embodiment of the invention, presenting a reduced coding cost, the whole is transmitted and / or stored on a data carrier motion vectors associated with the vertices of the settlement zone considered. The relevance of the motion vectors with regard to the overall motion estimation of the image is then evaluated at the time of decoding of the image. 3.2. Construction of a movement relevance tree

The "relevant" marker can take a TRUE value, when the movement of a vertex is relevant for the estimation of overall movement of the image, and must therefore be transmitted and / or stored on a data medium, or FALSE otherwise. We build the relevance tree from the faces and vertices of the hierarchical mesh belonging to a level of the pyramid of mesh less than or equal to L _m .

We define a relevance criterion, noted CP (T), with value in [0; 1], and a relevance threshold SP, also with value in [0; lj. It is of course conceivable that the relevance criterion and the relevance threshold are of value in any other interval [A; B] adapted to the invention.

For a given triangle T of the triangular mesh considered, the value of the "relevant" marker is defined as follows:

RELEVANT (T) = CP (T)> SP In addition, the criterion CP is defined by weighted combination of three value-based sub-criteria in [0; 1]: a surface CS sub-criterion allowing the relationship between the surface of the triangle T considered and the average surface of the triangles of the same level of the mesh: CS (T) = surf (T) / avg {surf (T ') such that T belongs to the same level of mesh as T}; a sub-criterion CC of compactness, possibly standardized, making it possible to evaluate the relationship between the surface and the perimeter of the triangle T considered: CC (T) = (1 / CC (Equilateral) * surf (T) / [4 * π * perimeter (T) ² ]. Such a criterion is maximum and equal to 1 if T is equilateral, and it is zero if T is flat; a sub-criterion CA of antagonism of the movements of the vertices of the triangle considered. Such a criterion is worth 1 if the movements of the vertices of T are antagonistic, that is to say in the presence of a potential mesh reversal, and 0 if they are of the same direction and direction. If Dl, D2 and D3 are the motion vectors of the vertices of T, then:

CA (T) = 1 if sign (<D1, D2>) <0 or sign (<D3, D2>) <0 or sign (<D1, D3>) <0 CA (T) = 0 otherwise.

Thus, CP (T) = [a * CS (T) + b * CC (T) + c * CA (T)] / [a + b + c] where a, b, and c are predetermined scalars.

We detail below an example of an algorithm that can be implemented to assign the "relevant" marker to the vertices and triangles of the mesh, and therefore determine the motion vectors to be transmitted and / or stored on a data carrier: all the vertices S of the mesh pyramid traversed from the levels L ₀ to L _m

The "relevant" marker of S is initialized to FALSE For all the triangles T of the mesh pyramid traversed from the levels L ₀ to L _m If CP (T)> SP _m

Then the "relevant" marker of T takes the value TRUE the "relevant" marker of the vertices of T takes the value TRUE Otherwise the "relevant" marker of T takes the value FALSE For all the vertices S of the pyramid of mesh traversed by the levels L ₀ to L _m If "relevant" (S) then transmit the movement of S.

3.3. Management of mesh occultations

When, for example, two objects intersect within the image, occultation phenomena can occur, which can be managed according to two distinct methods according to the invention. 3.3.1. Merger of vertices The first method implemented according to the invention consists in detecting, at the time of decoding, the occultation zones determined by the faces 130, 133 of the mesh whose vertices 131, 132 have antagonistic motion vectors, according to the sub- CA criterion, as illustrated in FIG. 13. Indeed, the faces 130, 133 thus detected are liable to turn over because their vertices 131, 132 are positioned on different objects, one of the two objects obscuring its neighbor.

According to the particular embodiment described here, an edge fusion (in English "edge collapse") is then practiced between the two neighboring vertices 131, 132 having an antagonistic movement. This results in the disappearance of a triangle 130, 133, reflecting the disappearance of part of an object. The vertex obtained by merging the vertices 131 and 132 is placed on the edge initially connecting the vertices 131 and 132, for example in the middle, or at the position of one of the two vertices 131 and 132, as illustrated in Figure 13, where the merged vertex is located in the position initially occupied by the vertex 132. 3.3.2. N-manifold mesh A second method implemented according to the invention for managing the occultation phenomena of meshes consists in working on an n-manifold mesh, as illustrated in FIG. 14. An n-manifold mesh is defined as a triangular mesh in which an edge can be shared by n triangles, where n≥2.

Thus, FIG. 14 presents an example of a 3-manifold mesh, in which the edge referenced 143 is common to the three triangles referenced 140, 141, and 142. According to such a method, an motion is first estimated. of the image, without taking into account the "relevant" marker described above. At the time of the decoding of the image, when an occultation zone (that is to say a reversal of a triangle of the mesh considered) is detected, the triangles associated with this zone then become carriers of a marker "overlapped" (or in French "overlapped"). We then proceed to a new optimization of the movement of the image, by excluding the triangles of the mesh carrying the marker "overlapped".

This new optimization can lead to the appearance of new mesh reversals, which in turn carry a "overlapped" marker. We then proceed to a new optimization of the movement.

Thus, the zones of the mesh carrying the "overlapped" marker mark the discoveries or the overlaps of meshes: they therefore correspond to occulted objects.

The method implemented in the particular embodiment described in this paragraph consists in temporarily removing the triangles carrying the "overlapped" marker, while keeping them in memory, so as to be able to manage their possible reappearance on the image.

According to the topology of a mesh area carrying the "overlapped" marker (also called "overlapped" sub-mesh), the two cases are distinguished below: the boundaries of the "overlapped" sub-mesh are rescinded, and the mesh then becomes n-manifold (3-manifold in the example in Figure 14); only part of the borders of the sub-mesh is rescindée, and one operates a local correction of the contents of the other meshes "overlapped". The use of an n-manifold mesh advantageously makes it possible to conserve the photometric and / or colorimetric information associated with areas of the mesh which may disappear or appear repeatedly during the video sequence considered.

3.3. Conclusion The introduction of a criterion of relevance of the movement makes it possible to reduce the cost of the transmission and / or storage of the vectors of movement of the settlement zones, while preserving the appearance of the reversals of meshes, necessary for the setting up. mesh content day. Thus, the estimation of the movement and the transmission and / or storage of the motion vectors are governed by a relevance tree, which makes it possible to ensure the quasi-consistency of the transmission rate / image distortion ratio, despite the entry of new meshes in the image, and the settlement of some others.

Claims

1. Method for coding a mesh representative of an image of an animated sequence, characterized in that said mesh is of size greater than said image.

2. Coding method according to claim 1, characterized in that, said image having an area of N * M pixels, said mesh has an area of at least nN * nM pixels, n being an integer greater than or equal to 2.

3. Coding method according to any one of claims 1 and 2, characterized in that said mesh is a hierarchical mesh having at least two levels of mesh nested within a pyramid of mesh.

4. Coding method according to any one of claims 1 to 3, characterized in that said mesh is a triangular mesh consisting of an arrangement of vertices and triangular faces, each defined by three references to the vertices which it connects, and having three edges each connecting two of said vertices.

5. Coding method according to any one of claims 1 to 4, characterized in that it associates with at least some of the vertices and / or some of the faces of said mesh at least one of the following markers: an "internal" marker, a face carrying an "interior" marker if it has a non-empty intersection with said image, and a vertex carrying an "interior" marker if it belongs to a face

"interior"; an "incoming" marker, a vertex carrying an "incoming" marker if it has an "inside" marker and is outside the image, and a face carrying an "incoming" marker if at least one of the three vertices that it links carries an "incoming" marker.

6. Coding method according to claim 5, characterized in that it implements the following succession of steps: estimation of the movement of said vertices of said mesh, between two successive images of said sequence; construction of colorimetric and / or photometric information intended for vertices and / or faces of said mesh carrying said "incoming" marker.

7. Coding method according to claim 6, characterized in that said estimation step comprises the following sub-steps: optimization of motion vectors of the vertices of said mesh carrying said "interior" marker, so as to minimize a criterion of predetermined error in reconstruction of the following image; interpolation and / or extrapolation of motion vectors from the vertices of said mesh not carrying said "interior" marker, in order to fluidize the overall displacement of said mesh.

8. Coding method according to claim 7, characterized in that said interpolation and / or extrapolation step implements a double iterative front-back scan, applied to the vertices of the lowest level of said mesh pyramid for which said motion vectors of said vertices have been optimized.

9. Coding method according to any one of claims 7 and 8, characterized in that after said estimation step, it implements a step of displacement of each of the vertices of said mesh, by applying to it its own vector of movement.

10. Coding method according to any one of claims 7 to 9, characterized in that said construction step implements on the one hand a finite element coding model, and on the other hand an alternative coding model, the latter being implemented for at least certain areas of said image, according to a predetermined error criterion.

11. Coding method according to any one of claims 7 to 10, characterized in that the said construction step implementing a hierarchical coding model, it comprises the following substeps: normal temporary projection of said vertices carrying said "incoming" marker on an edge of said image; optimization of said colorimetric and / or photometric information of said vertices and or of said faces of said image, so as to minimize a predetermined error criterion for reconstruction of the following image; repositioning of said vertices carrying said "incoming" marker to their position before projection, said "incoming" vertices carrying said optimized colorimetric and / or photometric information.

12. Coding method according to claim 10, characterized in that said specific coding comprises the following substeps: construction of a symmetrized image, obtained by applying to said image an axial symmetry with respect to each of its edges and a symmetry central to each of its vertices; optimization of said colorimetric and / or photometric information of said vertices and / or of said faces of said symmetrized image by application of said alternative model (DCT, fractals, matching pursuit).

13. Coding method according to any one of claims 1 to 12, characterized in that it implements a step of memorizing photometric and / or colorimetric information associated with the vertices and / or the faces emerging from said image.

14. Coding method according to any one of claims 6 to 13, characterized in that, as long as one face of said mesh carries the "incoming" marker, said construction step implements at least one iteration of the substeps successive steps: - optimization of said colorimetric and / or photometric information of said vertices of said "incoming" face, so as to minimize a predetermined error criterion; transmission to a terminal and / or storage on a data medium of said optimized colorimetric and / or photometric information.

15. Coding method according to any one of claims 6 to 14, characterized in that, as long as one face of said mesh carries the "incoming" marker, said construction step implements at least one iteration of the sub-steps successive steps: - elaboration of a criterion making it possible to evaluate an image reconstruction quality obtained by taking into account the current photometric and or colorimetric information of the vertices of said face; when said quality is not satisfactory: optimization of said colorimetric and / or photometric information of said vertices of said face, so as to minimize a predetermined error criterion for reconstruction of the following image; transmission to a terminal and / or storage on a data medium of said optimized colorimetric and / or photometric information; transmission and / or storage of a tree of markers indicating to said terminal whether said colorimetric and / or photometric information of said vertices of said "incoming" face have been optimized or not.

16. A coding method according to any one of claims 6 to 15, characterized in that said movement estimation step further comprises a substep of construction of a tree of motion vectors to be transmitted and / or to store, the belonging of a motion vector to said tree being determined from a criterion of relevance of the movement applied to said faces of said mesh.

17. Coding method according to claim 16, characterized in that said relevance criterion results from a predetermined weighted combination of the following sub-criteria: a surface sub-criterion, making it possible to evaluate the ratio between the surface of a face and the average surface of the faces of said mesh belonging to the same level of said mesh pyramid; a compactness sub-criterion, making it possible to evaluate the relationship between the surface and the perimeter of a face of said mesh; a sub-criterion of antagonism of the movements, making it possible to evaluate the antagonism of the motion vectors of vertices of said mesh.

18. Coding method according to any one of claims 6 to 15, characterized in that all the motion vectors of said vertices of said mesh are transmitted and / or stored, a suitable decoder identifying according to a predetermined criterion the vectors to be taken into account. account.

19. Coding method according to any one of claims 1 to 18, characterized in that in the presence of an occultation of at least one vertex occulted by a face comprising at least one occulting vertex, said fusion is carried out. occulted vertex with said occulting vertex.

20. Coding method according to any one of claims 1 to 19, characterized in that it implements a step of detecting at least one area of said obscured image during a displacement of said mesh, and a step of memorizing photometric and / or colorimetric information relating to said obscured area, in view of possible future use, said information not being processed.

21. A method of decoding a mesh representative of an image of an animated sequence, characterized in that said mesh is of size greater than said image.

22. Signal representative of a mesh representative of an image of an animated sequence, characterized in that said mesh is of size greater than said image.

23. Device for coding a mesh representative of an image of an animated sequence, characterized in that said mesh is larger than said image.

24. Device for decoding a mesh representative of an image of an animated sequence, characterized in that said mesh is larger than said image.