FR2811112A1 - Encryption of 2D or 3D image sources for adaptive viewing and method of decryption, generates a base binary train of image region and then a binary refining train independently processed to produce a better image - Google Patents

Abstract

A region of interest is selected by an operator (104). Each of the regions of interest is associated with at least a rectangular zone defined at the heart of the image source by the operator. Each rectangular zone includes the associated region of interest. An Independent Claim is included for - an encrypting device. Method for encrypting an image source (101) using a tree structure to represent an hierarchical mesh associated with the image generate a base binary train (103) representing a global structure of the mesh (102) which allows reconstruction of a close representation of the image source; select (105) a region of interest at the heart of the image; generate a localizing binary train (107) which allows localization of the region(s) of interest at the heart of the image; generate a refinement binary train (109) representing each of the regions of interest, each of trains can be independently processed and allows reconstruction of a refined representation (108) of the region of interest.

Description

1 2811112

  Method for coding a source image with adaptive display,

  decoding method and corresponding decoder.

  The field of the invention is that of coding of fixed or moving images.

  More specifically, the invention relates to the progressive transmission of data relating to the images. In the field of image processing and transmission, there is an increasing interest in the progressive transmission of data, to alleviate in particular the problems of bandwidth saturation and

  adaptability to terminals of various kinds.

  The invention can be applied to 2D images, for example to allow a user wishing to view images comprising the objects of his wish, to download all of these images in small format, to select areas of interest from these thumbnails, and then zoom in on these areas in order to

view in detail.

  The invention also applies to 3D images, so as to allow detailed visualization of visually relevant areas of 3D objects, such as

  regions close from the point of view of an observer for example.

  Several image coding and compression techniques are known, such as the standard MPEG1 / 2 techniques, implementing compression

frame to frame data.

  A disadvantage of these techniques of the prior art is that they do not allow interactivity based on the content of the image, that is to say that they do not allow, in particular, to adapt to the requests of '' a user wishing, for example, to view certain areas of interest of an image with a

  higher quality level, or zoom in on certain regions of an image.

  However, recent interactive multimedia developments have shown the need to manipulate, compress, index video objects whether they are of natural or synthetic origin, to perceive areas of interest in an image, etc.

2 2811112

  We therefore sought to improve coding and compression techniques

  classics to make them interactive.

  However, to represent areas of interest of an image, it is necessary to code, on the one hand, their shape, and on the other hand, their texture. We therefore defined transparency plans for the MPEG4 standard and important work, relating to the corresponding coding, was published by J. Ostermann and FW Meier in the document "MPEG4 and rate distortion based shaped coding techniques" ("MPEG4 and suitable coding techniques based on the rate of

  distortion ", Proceedings of the IEEE, 86 (6): 1126-1154, June 98); in particular.

  a substantial bibliography of coding with or without loss of information from

  outline and texture has been developed.

  A disadvantage of this technique of the prior art is that the coding of such transparency plans represents a significant additional cost to the techniques.

classics based frames.

  According to an image coding technique using a bit plan approach, to code a particular form of an object, a binary mask is applied to the image, by associating the value 1 with all the points (or

  pixels) of the object, and the value 0 to the rest of the image.

  One then transmits macro blocks (for example of dimension 16 * 16), preceded by information indicative of belonging to a border, of the transparent character (for the object) or opaque (for the rest of the image) of the macro block. The shape of the object is then coded (for example according to a SADCT type coding), and information relating to a rectangular area is then transmitted.

  in which it is included (in English "bounding box").

  According to a classic technique, alphas planes (or transparency planes) are obtained by a color inlay technique (in English "chromakey"), allowing precise and rapid segmentation of an image into areas of interest, but at the cost specific shooting conditions (it is particularly

  necessary that the image be constructed on a blue background, blue being a color-

3 2811112

  key). For video source objects in frame mode, automatic extraction methods by spatio-temporal segmentation have been proposed, but the difficulty of automatically extracting objects significant from a semantic point of view is still current as well as illustrated by the document of A. M. Tekalp, entitled "Two dimensional mesh based visual object representation for interactive synthetic / natural digital video" ("Representation of a visual object based on a two-dimensional mesh for interactive video

  synthetic / natural ", Proceedings of the IEEE 86 (6): 1029-1051, June 98).

  The invention particularly aims to overcome these drawbacks of the art

prior.

  More specifically, an objective of the invention is to provide a method of

  image coding suitable for progressive data transmission.

  Another objective of the invention is to implement methods of

  coding and decoding allowing adaptive visualization of images.

  Another object of the invention is to provide coding and decoding methods allowing a partial reconstruction of the images, adapted to the request.

of a user.

  Another object of the invention is to implement methods of coding and decoding images which make it possible to overcome the problems of saturation.

  of bandwidth and adaptability to terminals of various capacities.

  Another object of the invention is to provide coding and decoding methods making it possible to generate an image combining different levels of quality. Yet another objective of the invention is to implement methods of coding and decoding images making it possible to selectively transmit to the display terminal the data corresponding to the wishes of a user.

  and / or associated network-terminal constraints.

  Another object of the invention is to provide coding and decoding methods which are simple and inexpensive to implement. In particular, an objective of the invention is to implement coding and decoding methods

4 2811112

  allowing an image to be reconstructed by areas of interest, without generating

  significant additional cost in computing time.

  These objectives, as well as others which will appear subsequently, are achieved using a method of coding a source image using a tree representative of a hierarchical mesh associated with said image. According to the invention, such a method comprises the following steps generation of a basic bit stream representative of a global structure of said mesh, making it possible to reconstruct an approximate representation of said source image; - selection of at least one region of interest within said source image; generation of a location bit stream, making it possible to locate said region or regions of interest within said source image; generation of a binary refinement train representative of each of said regions of interest, each of said binary refinement trains being able to be processed independently and making it possible to reconstruct a

  refined representation of said region of interest.

  Thus, the invention is based on a completely new and inventive approach to coding an image with a view to its transmission. Indeed, the invention is based in particular on the splitting of the binary data stream associated with an image into several binary trains, themselves scalable (in English "scalable" or

  "scale able") individually, and linked by a common structure.

  Thus, the invention implements an interactive segmentation of the regions of interest and does not generate any significant additional cost in computing time. In addition, the set of regions to be coded being based on a common structure, transmitted by the basic bitstream, it is no longer necessary, unlike the techniques

  of the prior art, to transmit information relating to the outline of the objects.

  Finally, the existence of a common structure allows a user to himself select regions of interest within the image viewed on a terminal, without

  that these regions have been specifically coded by a server beforehand.

2811112

  According to a preferred embodiment of the invention, said at least

  a region of interest is selected by an operator.

  Such an operator is conventionally a graphic designer, who selects the regions of interest during an image pre-processing step. Such regions of interest can be particular objects of the image, groups of objects located in the same visual plane, the background of the image, etc. According to an advantageous technique of the invention, each of said regions of interest is associated with at least one rectangular area, defined within said source image by an operator, each rectangular area including said region

of associated interest.

  Thus, it is not necessary to transmit information making it possible to define the outline of a region of interest, the shape of which can be complex. Each region of interest is included in a rectangular area, which makes it possible, during the decoding of the bit streams, to process a region of interest in a simple and precise manner, being able to decode such a region of interest thanks to the rectangular area

  associated, which indicates the binary train to choose.

  Advantageously, said location bit stream contains at least some of the following information: - the number of rectangular areas defined within said source image; - information making it possible to locate each of said rectangular areas in a coordinate system associated with said source image; - for each of said rectangular areas, an identifier of said region

of associated interest.

  At the time of decoding, it is thus easy to locate the rectangular areas within the image, from the coordinates of their vertices in a coordinate system associated with the image, or, for example, from the coordinates of a vertex, and knowledge of the width and length of the rectangular area considered. In addition, the identifier transmitted makes it possible to know which region of interest preselected by a graphic designer is each associated

rectangular areas.

6 2811 112

  Some rectangular areas may overlap. When at the time of decoding, the domain of the image which one wishes to decode (for example, a point of interest of an observer) is located on a zone of intersection of several rectangular zones, one then decodes at at least partially the binary refinement trains associated with these zones. According to an advantageous technique, the coding of each of said regions

  of interest includes hierarchical coding.

  Advantageously, the coding of each of said regions of interest is a hybrid coding, selectively implementing a coding by DCT and a coding

by finite elements.

  Preferably, said basic bit stream contains at least some of the following data: information representative of the number of meshes of said mesh; - information representative of the number of levels of said tree representative of said mesh; - the type of mesh of said mesh; - the nature of said mesh;

  - basic information of said source image.

  Thus, one can for example indicate the initial number of meshes of the mesh by indicating the number of points to be inserted by lines and by columns to generate it. One can also indicate, in the basic binary train, if the mesh consists of triangular, quadrangular meshes, etc. as well as the shape of the mesh (staggered mesh, square mesh, etc.). The basic information of the source image corresponds in particular to the low frequencies of the coding of

the image.

  According to a first advantageous variant of the invention, when a coarse region of interest includes at least one region of interest to be detailed, said bit stream of refinement of said coarse region of interest contains coarse information on said at least one region of interest to be detailed, and said bit stream of refinement of said at least one region of interest to be detailed does not

7 2811112

  contains only refinement information on said region or regions of interest to be detailed, so as to avoid transmission of redundant information. This avoids a redundancy of the information conveyed by the binary refinement trains. The refinement binary train of the region of interest to be detailed only corresponds to a sequence of refinements of the information

  coarse carried by the binary train of refinement of the encompassing region.

  It is of course also possible to adopt a simpler coding technique, consisting in authorizing the transmission of redundant information at the level of

encompassing regions of interest.

  According to a second advantageous variant of the invention, when an encompassing region of interest encompasses at least one contained region of interest, said bit stream of refinement of said encompassing region of interest does not contain information on said at least a region of interest contained, but contains at

  at least one identifier of said at least one region of interest contained.

  Thus, the refinement bit stream of the encompassing region of interest only contains information relating to the region complementary to the contained region of interest with respect to the encompassing region of interest. The refinement binary train of the region of interest contained contains all the information

  necessary for the visual restitution of the region contained.

  In this way, when the user decides to decode the surrounding region, he can access the partial content of the contained region, that is to say the content

  coarse at the same level as the surrounding region.

  The invention also relates to a method of decoding a coded image from a source image, a hierarchical mesh being associated with said

source image.

  According to the invention, such a decoding method implements at least one of the following steps:

8 2811 112

  - reception of a basic bit stream, representative of a global structure of said mesh, making it possible to reconstruct an approximate representation of said source image; - reception of a location bit stream, making it possible to locate at least one region of interest within said source image; - reception of a request identifying a domain to be decoded from said coded image, coming from a terminal; at least partial reception of at least one bit stream of refinement corresponding to said at least one region of interest of said source image, said at least one region of interest at least partially coinciding with said domain to be decoded; - at least partial processing of said at least one binary refinement train at least partially received, so as to reconstruct at least

  partially said domain to be decoded.

  Thus, the invention is based on a completely new and inventive approach to decoding data for the reconstruction of an image or a sequence of images. Indeed, the invention is based in particular on the reception, at least partially, of a plurality of binary trains conveying information characteristic of the image to be constructed, the decoding of a basic binary train making it possible to synthesize a structure common to the regions of interest defined within the image and of generating an image of coarse quality and / or of reduced size, and the decoding of a plurality of binary streams of refinement making it possible to determine the data of luminance, chrominance, etc. , associated with each of the regions of interest in the image. The basic bitstream in particular carries basic information, for example, relating to chrominance (UV) and to the

luminance (Y) of the image.

  Thus, the reception of one or more bit streams of refinement makes it possible to reconstruct certain areas of the image with a greater or lesser level of detail, depending on whether the associated bit stream has been received in its entirety or

partially.

9 2811 112

  The basic, location, and refinement binary trains can in particular come from a server (for example a server of an internet type network), and / or from a storage area, such as a CD ROM, a memory area of a suitable terminal, etc. Advantageously, said reception of said at least one refinement bit stream is a function of a transmission capacity of said coded image, and said processing of said at least one received refinement bit stream is a function of processing capacity and / or a screen capacity and / or a capacity of

  memory of a terminal on which said decoding method is implemented.

  Indeed, certain constraints of the transmission network, such as bandwidth constraints, can prevent the transmission of all the bit streams of refinement associated with an image. Binary trains are then only partially received, depending on the allocated bandwidth. Similarly, certain binary streams of refinement can only be partially decoded, depending on the storage capacity of the image viewing terminal, or

  depending on the quality of its screen, for example.

  Such a decoding method according to the invention thus allows an adaptive display of the images, according to the constraints of the terminal network. In particular, such a decoding method makes it possible to adapt to terminal types and therefore

of varying capacities.

  According to an advantageous characteristic of the invention, said domain to be decoded is selected by a user of said terminal and / or as a function of its distance from the eye of said user, the domain closest to the eye of said user

being decoded as a priority.

  Decoding can thus be implemented according to a request from a user, who selects the area of the image he wishes to view as a priority, or who zooms in on part of the image he wishes to view in details. Decoding can also be implemented according to a request from a virtual camera, associated for example with the eye of an observer of the image. We

2811112

  then determines the regions of the image that are closest to the camera, also called hot spots, so as to decode and view them with a high level of quality. Binary trains associated with regions farther from the camera, also called cold spots, are more coarsely decoded, and those associated with regions invisible to an observer are not decoded. Preferably, the processing of a region of interest is accompanied by the at least partial processing of at least one visible neighboring region, so as to avoid

  significant quality differences between neighboring regions.

  Indeed, it is important to ensure visual continuity within the image. Thus, when the decoding binary train of a region of interest is decoded with a high level of detail, the binary trains associated with the visible neighboring regions are also decoded, with a lower level of quality, so as to implement a gradual deterioration in the level of detail,

  region of main interest to peripheral regions of interest.

  According to an advantageous embodiment of the invention, on receipt of said basic binary train, a tree representative of said mesh is generated, and, during said processing of said at least one refinement binary train, we store, in said tree, information contained in said binary train (s)

refinement treated.

  The constitution of such a tree thus allows a decoding of the data by level, the tree being traversed in depth when one wishes to visualize a

  region of interest with a high level of detail.

  Advantageously, when said memory capacity allows it, information contained in said already received bit streams is stored in said tree representative of said mesh, so as to have a maximum

  information concerning said coded image.

  The data contained in the binary trains decoded in the tree is thus progressively stored, so that it can be accessed later in response to a camera request or a user request. If the memory capacity of the terminal makes it possible to store all of the information relating to an image, it is thus possible to respond to any camera request or to any request from a user, without implementing a new

  decoding of binary trains associated with the image.

  According to an advantageous technique of the invention, said memory capacity not making it possible to store all the information contained in said received bit streams, it is chosen not to store information associated with an area of said coded image not visible to said user, then possibly to areas located at a distance from the eye of said user, so as to store in priority in said tree information associated with the area of

  said coded image closest to the eye of said user.

  Indeed, the memory capacity of the display terminal may be insufficient to store all of the information carried by the binary trains associated with an image. It is then necessary to gradually erase the unnecessary information, so as to keep only the information relating to a current request. The information associated with areas invisible to an observer is then erased as a priority, then, if necessary, areas that are barely visible or

  very far from the camera and / or the eye of an observer.

  The invention also relates to a coding device implementing a coding method as described above, and a display terminal and a

  decoding device implementing a decoding method as described above

  above. According to the invention, within such a display terminal, the image data to be displayed are developed in video memory only at the time of

their display.

  The invention also relates to a signal structure representative of a source image coded according to the coding method and / or decoded according to the decoding method described above. According to the invention, such a signal structure comprises: - a basic bit stream representative of a global structure of a hierarchical mesh associated with said source image, and making it possible to reconstruct an approximate representation of said source image;

12 2811112

  a location bit stream, making it possible to locate at least one region of interest previously selected within said source image; - for each of said regions of interest, a binary refinement train, which can be processed independently and making it possible to reconstruct a refined representation of said region of interest. The invention also relates to a system for transmitting a mesh representative of a source image, between, on the one hand, at least one coder and / or at least one data carrier, and on the other hand, at least one display terminal, such a system implementing an encoding method and / or a

  decoding method as described above.

  The invention can advantageously be applied to one of the following operations: - data transmission over a network of the internet and / or intranet type; - visualization of images relating to cultural applications; - visualization of images and / or sequences of images relating to an interactive game; - transmission and / or visualization of data relating to collaborative work;

- telecommerce.

  The invention also relates to a data medium and a server of a communication network, comprising at least one signal representative of a source image encoded according to an encoding method as described above. Such a server can in particular be a server of the global Internet network, or of any

other internet type network.

  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 appended drawings, among which: - Figures 1 to 3 show examples of selection of regions of interest within a source image;

13 2811112

  FIG. 4 illustrates the preparation of the data relating to the regions of interest selected in FIGS. 1 to 3, before the implementation of the coding method according to the invention; FIG. 5 presents an example of a camera request, making it possible to determine the regions of interest of FIGS. 1 to 3 to be decoded; - Figure 6 illustrates an example of a nested hierarchical mesh; - Figure 7 shows an example of coding rectangular areas in a location bit stream; - Figure 8 illustrates an example of a user request to zoom in on part of the image; - Figure 9 shows an example of a pyramid of meshes representative of the image, as well as the associated tree;

  - Figure 10 is a simplified block diagram recalling the principle of the invention.

  The general principle of the invention is based on the interactive segmentation of an image into areas of interest, and on the coding of the data relating to each of the areas of interest in an independent binary train, so as to allow a selective reconstruction of certain regions of the image, based on a request

  camera and / or user request.

  This general principle is illustrated in FIG. 10. We extract from a source image 101, with which a mesh is associated, the information characteristic of the global structure 102 of the mesh. This information is conveyed by a basic binary train 103, which makes it possible to reconstruct an approximate representation of

source image 101.

  Under the action 104 of a user at the level of the display terminal, or for example of a graphic designer at the level of the server, one or more regions of interest are selected within the image 101 during a step referenced 105. Thus, at the display terminal, a user can send a request to the server to receive the binary trains corresponding to a region of the image, selected for example using a

14 2811112

  cursor. A given positioning of the user's cursor is associated with a pixel contained in a box, whose index is determined, in order to identify the train (s)

  binary (s) that it is necessary to transmit to the user in response to his request.

  The selection of the regions of interest during the step referenced 105 can also be implemented by a graphic designer at the server level. The graphic designer then prepares the data necessary for generating the trains.

  binaries that can be sent in response to a user request.

  The regions of interest selected are located 106 within the source image, and the number of the box containing them is conveyed by a binary train of

location 107.

  Refinement information, making it possible to reconstruct a refined representation 108 of the regions of interest, is then extracted from each of the selected regions of interest, and conveyed by a plurality of corresponding bit streams of refinement 109. Each refinement bit stream is associated with a region of interest of the source image 101 and can be processed independently. We present, in relation to FIGS. 1 to 3, examples of selection of regions of interest within a source image. We consider images on which pre-treatments are carried out, with a view to implementing the method.

  adaptive local decoding with progressive transmission according to the invention.

  We are interested in image 11 of figure 1. We chose on this image 11 to select the objects referenced 12 and 13. We can notice, on the right of image 11, a series of squares, whose colors vary from black (which corresponds to a rough reconstruction quality) to white (which corresponds to a fine reconstruction quality). The objects referenced 12 and 13 are shown in white on the thumbnail 15, so as to indicate that one wishes to view them with a

  maximum reconstruction quality.

  FIG. 1 therefore illustrates an example of possible application of the invention, consisting in selecting all the objects contained in an image (the background of the image can be considered as a particular object), and in associating with

2811112

  each of these objects has a reconstruction quality and / or a given transmission rate. FIG. 2 illustrates the selection, within the source image 21, of two regions of interest 23 and 24, to which it is desired to assign different qualities of reconstruction and / or different transmission rates. Thus, it can be seen on the thumbnail 22, in relation to the series of squares 25, that the object 24 must be viewed with a quality of fine reconstruction, while the object 23 can be

  viewed with a coarser reconstruction quality.

  It can also be noted in image 21 that the regions of interest 23 and

24 are slightly nested.

  We now present, in relation to Figure 3, the selection of six

  regions of interest nested within a 3D image texture 31.

  The thumbnail 32 indicates that four distinct levels of reconstruction quality are required for image 31: regions 35 to 38 are the most visually relevant regions, as they correspond to the eyes, nose, ears and mouth of the face 31, and must therefore be reconstructed with a fine reconstruction quality. The region of interest 37 is associated with a lower quality of reconstruction. The region of interest 33 corresponds to an even coarser quality, and the background of the image 31 is associated with the quality of reconstruction.

minimal.

  Such a determination of the areas of interest of an image, as illustrated in FIGS. 1 to 3, is particularly advantageous for graphic designers, in

  view of a dependent application from a user's point of view.

  We now consider, in relation to FIGS. 4 to 7, the techniques of approximation and coding of the areas of interest selected in FIGS. 1 to

  3, implemented in an image coding method according to the invention.

  In a particular embodiment of the invention, a preliminary step to the coding method consists in determining a structure common to all of the regions of interest selected. In this way, the subsequent implementation of the decoding method according to the invention depends only on the requests

16 2811112

  of a user. In particular, it is then possible to decode, as desired, the bit streams generated during coding, and to reconstruct an image comprising the areas of interest chosen by the user, with the level of quality of reconstruction

desired by the latter.

  Each region of interest is then approached and coded independently. To the bit stream associated with a given region of interest, information is also added which makes it possible, during decoding, to identify the region of interest considered, without it however being necessary to code (and therefore to

  decode) the explicit form of this region.

  There is presented, in relation to FIG. 4, an embodiment consisting in constructing, around the regions of interest selected, rectangular zones including them. Throughout the rest of the document, such rectangular areas are

called box.

  There are four regions of interest R1 to R4 within image 40. We therefore construct three boxes referenced 41 to 43 including the regions of interest R1 to R3 respectively. The region of interest R4 is, for its part, limited by the frame of the image 40. It can also be seen that the boxes 41 and 42 partially overlap. The construction of such boxes makes it possible in particular, during decoding, to reconstruct the image 40 according to a camera request, or a request from a user. Such a notion of camera request is illustrated in Figures Sa and 5b. Figure Sa presents a three-dimensional object 51, the associated texture of which is

illustrated in Figure 5b.

  Zone 52 is the zone closest to a virtual camera which would observe the object 51. Zone 52 therefore corresponds to a hot spot of the camera, and must therefore be decoded precisely. On the other hand, the zone 53 corresponds to a cold point of the camera, and can be coarsely decoded. Zone 54

  is not visible from the camera and will therefore not be decoded.

17 2811112

  To optimize the content of each region of interest selected, we then use an approximation technique similar, for example, to

  techniques described in patent applications FR 98 12525 and FR 99 06815.

  The first step of the coding method according to the invention then consists in generating the bit stream associated with the coding of the structure common to the set

the regions.

  The structure common to the regions of interest selected corresponds to a nested regular hierarchical triangular mesh 61, as illustrated in FIG. 6. It is therefore necessary, when coding this structure, to generate a basic bit stream comprising for example the number of initial meshes, the number of levels of the pyramid of mesh, the type of meshes (for example, triangular, quadrangular, etc.) and the type of mesh (for example, staggered mesh, square mesh, etc.). According to a preferred embodiment, the number of initial meshes is coded, in the basic bitstream, in the form of the number

  mesh points to insert by row and by column.

  The second step of the coding method according to the invention consists in coding, for each region of interest selected, all of the data

  necessary for its reconstruction, in a binary train of refinement.

  The technique used during this second step is similar, for example, to that described in French patent applications FR 98 12525 and FR 99 06815, for coding the luminance, chrominance, and possibly position data of the different meshes representative of the image considered. In addition to these data, the chosen quantifications are also coded, possibly the number of mesh levels, etc. according to a technique

  similar to that described in the above-mentioned French patent applications.

  According to a particular embodiment of the invention, first of all a binary train is generated comprising basic information of the image considered, in order to allow a display terminal to have an image of size identical to l original image but of rough quality or else

  of an undersampled image which will therefore seem more detailed than the first.

18 2811112

  For example, the image under consideration is coded with a very high compression rate, so as to allow a very brief visualization of the latter at the destination terminal. Then, for each region of interest in the image, we generate

  a binary train comprising all the data described above.

  The basic bitstream, described above, having previously transmitted information relating to the base layer of each of the regions of interest to the display terminal, the refinement bitstream associated with each of these regions therefore only carries relative data. to a sequence of refinements of the values of the base layer of the regions of interest, in

  function of the quantifiers of each region.

  Furthermore, according to the invention, a header is associated with each bit stream of refinement, comprising an integer between zero and the number of regions of interest selected, and making it possible to identify, during the process

  decoding, the area of interest with which the received refinement train is associated.

  However, it can happen that the regions of interest represent nested zones, as illustrated in FIG. 3. In this case, in order to avoid redundancy in the transmission of information, the encompassing regions, representing the coarsest regions, may contain basic information from interior regions. The interior regions then contain only information of

data refinement.

  We can also authorize the transmission of redundant information, and each refinement bit stream then carries all the refinement data necessary for the refined reconstruction of the region of interest to which

he is associated.

  To avoid the transmission of redundant information when the regions of interest represent nested areas, it can also be envisaged that the bit stream of refinement of an encompassing region of interest does not contain

  information about a contained region of interest.

  A third step of the coding method according to the invention consists in generating a location bit stream, independent of the data relating to

19 2811112

  the source image, which contains information relating to the coordinates of each of the boxes described above, in a reference frame relating to the image considered. Such a location bit stream conveys for example the number of boxes relating to the image, their coordinates (or the coordinates of a vertex, the width and the length of each box), and the number of the region of interest that 'they

  contain, as shown in Figure 7.

  Image 71 is segmented into four boxes B I to B4. For each box Bi, therefore, in the location bit stream, the index i and the positions Pl, i, P2.i, P31 and P4i of its vertices are coded. It is also possible to code, in the location bit stream, the index i, one of the four positions Pj., And the values of the width and

the length of the Bi box.

  We now present, in relation to FIGS. 8 and 9, an example of implementation of the decoding method according to the invention. The decoding process of the bit streams described above is described in particular, according to the requests of a user and / or the problems encountered during the progressive transmission of the data, such as the saturation of the transmission network by

example.

  The various constraints relating to the decoding of the received bit streams may be bandwidth constraints, and / or processing and / or storage capacity of the display terminal, as well as constraints relating to

  a user request and / or a camera request.

  Thus, the user can, for example, during a request, request the transmission of a region of interest determined during coding to a more or less coarse precision. He can therefore decide whether the binary train associated with the region

  of selected interest must be transmitted in whole or in part.

  The user can also ask to zoom in on an area 82 common to several regions of interest of the image 81, as illustrated in FIG. 8.

  zone 81 intersects in particular the regions of interest defined by boxes B1 and B3.

2811112

  In this case, the regions located around the area 82 are partially decoded, in order to transmit only the information relating to the point of interest of the user, using the structure common to all of the regions of interest. The first step of such a decoding consists in decoding the binary train comprising the data associated with the common structure of the regions of interest, in order to resynthesize it. In the particular case where the common structure is represented by a pyramid of nested meshes 91, as illustrated in figure

  9, the decoder recreates these meshes and generates a tree 92.

  Each level 921 to 923 of the tree 92 corresponds to a level of the mesh 91, and each node of the tree 92 corresponds to a triangle of the mesh 91. We can therefore obtain a binary, tertiary, quaternary tree, etc. The second step consists in decoding the base layer of the image 81, in order to generate a thumbnail or an image of size equivalent to the image.

  original, but very rough quality.

  The third step then consists in decoding the bit stream associated with the

  box Bi and generate them fictitiously on image 81.

  During the fourth step of decoding, the first bits are decoded

  of each binary train in order to determine which box B1 to associate them with.

  Finally, the last step depends on camera requests and / or user requests, and / or bandwidth and / or capacities of the display terminal. Indeed, the user can issue a request by zooming in on a region of the image. Several configuration cases can then arise: - a first case is that where the region selected by the user corresponds to a predefined region of interest. This region being contained in a box, the box number is then used to decode the associated bit stream. If, during decoding, a problem with bandwidth (or terminal capacity) appears, the user will only see information appearing on the display terminal that could have been transmitted to the terminal (in fact, binary trains

21 2811112

  associated with the regions of interest correspond to sequences of refinements of

  the image). Finally, the decoded information is then stored in the tree.

  - a second case is that where the region selected by the user

  covers at least partially several predefined regions of interest.

  As shown in Figure 7, the boxes can overlap (for example, BI and B2 overlap). Therefore, the point of interest of the user can be located at the intersection of several boxes. We can then detect the regions of interest associated with the user's request and partially decode these regions according to the quality required by the user (linked for example to the importance of the

  zoom performed during the request).

  We can also consider a similar approach, which would be based on the request of the camera, and no longer on the request of the user. Indeed, we know that the hot spot of the camera corresponds to an integral decoding of the binary train associated with the selected region of interest, and that the more we move away from this point,

* the less data is decoded in the associated regions of interest.

  - A third case is that where all the regions of interest of the image 81 have been previously decoded. The entire data was then stored in a tree 92. Thus, depending on the orientation of the camera, it is possible to allow local adaptive visualization of the content of the image 81 (or texture in 3D) by performing a deep scan of the tree 92, in order to carry out the associated rendering

at the request of the camera.

  - a fourth case corresponds to the particular case of nested regions.

  Thus, it may happen that the camera zooms in on a region of interest included in an encompassing region. As the terminal has nesting information for the regions of interest (via the Bi boxes), the decoding process then consists in partially decoding the surrounding region, so as to recover all the relative information necessary for the reconstruction of the

region of interest.

22 2811112

Claims (25)

  1. Method for coding a source image (11; 21; 31; 40; 71; 81; 101) using a tree representative of a hierarchical mesh (61) associated with said image, characterized in that comprises the following steps: - generation of a basic bit stream (103) representative of a global structure of said mesh (102), making it possible to reconstruct an approximate representation of said source image; - selection (105) of at least one region of interest (12, 13; 23, 24; 33-38) within said source image; - generation of a location bit stream (107), making it possible to locate (106) said region or regions of interest within said source image; generation of a refinement binary train (109) representative of each of said regions of interest, each of said refinement binary trains being able to be processed independently and making it possible to reconstruct a   refined representation (108) of said region of interest.   2. Coding method according to claim 1, characterized in that said at   least one region of interest is selected by an operator (104).   3. Coding method according to any one of claims 1 and 2,   characterized in that each of said regions of interest is associated with at least one rectangular area (41, 42, 43), defined within said source image by an operator,   each rectangular area including said associated region of interest.   4. Method according to claim 3, characterized in that said location bit stream contains at least some of the following information: - the number of rectangular zones defined within said source image; - information making it possible to locate each of said rectangular areas in a coordinate system associated with said source image; - for each of said rectangular areas, an identifier of said region of associated interest. 23 2811112   5. Coding method according to any one of claims 1 to 4,   characterized in that the coding of each of said regions of interest comprises a hierarchical coding.   6. Coding method according to claim 5, characterized in that the coding of each of said regions of interest is a hybrid coding, putting in   selectively implements DCT coding and finite element coding.   7. Coding method according to any one of claims 1 to 6,   characterized in that said basic bit stream contains at least some of the following data: - information representative of the number of meshes of said mesh; - information representative of the number of levels of said tree representative of said mesh; - the type of mesh of said mesh; - the nature of said mesh; - basic information of said source image;   - quantifications associated with said source image.   8. Coding method according to any one of claims 1 to 7,   characterized in that, when a coarse region of interest includes at least one region of interest to be detailed, said refinement bit stream (109) of said coarse region of interest contains coarse information on said at least one region of interest to be detailed, and said refinement bit stream of said at least one region of interest to be detailed only contains refinement information on said at least one region of interest to be detailed,   so as to avoid transmission of redundant information.   9. Coding method according to any one of claims 1 to 7,   characterized in that, when an encompassing region of interest encompasses at least one contained region of interest, said bit stream of refinement of said encompassing region of interest does not contain information on said at least one region of interest contained, 24 2811112   and in that said refinement bit stream of said encompassing region of interest contains an identifier of said at least one contained region of interest,   so as to avoid transmission of redundant information.   10. Method for decoding a coded image from a source image, a hierarchical mesh being associated with said source image, characterized in that it implements at least one of the following steps: - reception of a train basic binary (103), representative of a global structure of said mesh (102), making it possible to reconstruct an approximate representation of said source image (101); - reception of a location bit stream (107), making it possible to locate (106) at least one region of interest within said source image; - reception of a request identifying a domain to be decoded (82) of said coded image, originating from a terminal; - at least partial reception of at least one bit stream of refinement (109) corresponding to said at least one region of interest of said source image, said at least one region of interest at least partially coinciding with said domain to be decoded; - at least partial processing of said at least one binary refinement train at least partially received, so as to reconstruct at least   partially said domain to be decoded.   11. A decoding method according to claim 10, characterized in that said reception of said at least one refinement bit stream is a function of a transmission capacity of said coded image, and in that said processing of said at least one bit stream of refinement received is a function of a processing capacity and / or a screen capacity and / or a memory capacity of a terminal on which said decoding method is implemented. 12. Decoding method according to claim 11, characterized in that said domain to be decoded is selected by a user of said terminal and / or by 2811112   depending on its distance to the eye of said user, the area closest to the eye   said user (52) being decoded as a priority.   13. A decoding method according to any one of claims 10 to 12,   characterized in that the processing of a region of interest is accompanied by the at least partial processing of at least one visible neighboring region, so as to avoid   significant quality differences between neighboring regions.   14. A decoding method according to any one of claims 10 to 13,   characterized in that upon reception of said basic bit stream, a tree (92) representative of said mesh is generated, and in that, during said processing of said at least one refinement bit stream, are stored, in said tree, information contained in said train (s) refined binaries processed.   15. A decoding method according to claims 11 and 14, characterized in that   that when said memory capacity allows it, information contained in said already received bit streams is stored in said tree representative of said mesh, so as to have a maximum of information concerning said coded image.   16. A decoding method according to any one of claims 14 and 15,   characterized in that, said memory capacity not making it possible to store all the information contained in said received bit streams, it is chosen not to store information associated with an area (54) of said coded image not visible to said user, then possibly to areas (53) located at a distance from the eye of said user, so as to store in priority in said tree information associated with the area of said coded image closest to the eye of said user.   17. Signal structure representative of a source image coded according to the method   of any one of claims 1 to 9 and / or decoded according to the method of   any one of claims 10 to 16,   characterized in that it comprises: 26 2811112   - a basic bit stream (103) representative of a global structure (102) of a hierarchical mesh (61) associated with said source image (101), and making it possible to reconstruct an approximate representation of said source image; - a location bit stream (107), making it possible to locate (106) at least one region of interest previously selected (105) within said source image; - for each of said regions of interest, a binary refinement train (109), which can be processed independently and making it possible to reconstruct   a refined representation (108) of said region of interest.   18. Transmission system for a mesh representative of a source image, between, on the one hand, at least one encoder and / or at least one data carrier, and, on the other hand, at least one display terminal , characterized in that it implements a coding method according to any one   of claims 1 to 9 and / or a decoding method according to any one of claims 10 to 16.   19. Display terminal implementing a decoding method according to   any one of claims 10 to 16, characterized in that the data   images to be displayed are developed in video memory only when their display.   20. Device for coding a source image (11; 21; 31; 40; 71; 81; 101) using a tree representative of a hierarchical mesh (61) associated with said image, characterized in that it implements the following means - means for generating a basic bit stream (103) representative of a global structure of said mesh (102), making it possible to reconstruct an approximate representation of said source image; - means for selecting (105) at least one region of interest (12, 13; 23, 24; 33-38) within said source image; 27 2811112   - means for generating a location bit stream (107), making it possible to locate (106) said region or regions of interest within said source image; means for generating a refinement binary train (109) representative of each of said regions of interest, each of said refinement binary trains being able to be processed independently and allowing   to reconstruct a refined representation (108) of said region of interest.   21. Device for decoding a coded image from a source image, a hierarchical mesh being associated with said source image, characterized in that it implements the following means: - means for receiving a train basic binary (103), representative of a global structure of said mesh (102), making it possible to reconstruct an approximate representation of said source image (101); - means for receiving a location bit stream (107), making it possible to locate (106) at least one region of interest within said source image; - means for receiving a request identifying a domain to be decoded (82) of said coded image, originating from a terminal; - Means for at least partial reception of at least one binary refinement train (109) corresponding to said at least one region of interest of said source image, said at least one region of interest at least partially coinciding with said domain to be decoded ; means for at least partial processing of said at least one binary train of refinement at least partially received, so as to reconstruct at   at least partially said domain to be decoded.   22. Applications of the coding method according to any one of   claims 1 to 9 and / or the decoding method according to any one of   Claims 10 to 16 in one of the following operations:   - data transmission over an internmet and / or intranet type network; - visualization of images relating to cultural applications; 28 2811112   - visualization of images and / or sequences of images relating to an interactive game; - transmission and / or visualization of data relating to collaborative work; - telecommerce. 23. Data carrier, characterized in that it comprises at least one signal representative of an image   coded source according to any one of claims 1 to 9.   24. Server of a communication network, characterized in that it comprises at least one signal representative of an image   coded source according to any one of claims 1 to 9.   25. Coding method according to any one of claims 1 to 9, and / or   decoding method according to any one of claims 10 to 16, and / or   signal structure according to claim 17, and / or transmission system according to claim 18, and / or display terminal according to claim 19, and / or coding device according to claim 20, and / or decoding device according to claim 21, and / or data medium according to claim 23, and / or server according to claim 24, characterized in that said basic bit stream contains basic information   of colorimetry and photometry and possibly the associated quantifications.