1 . Scope of the invention

The invention relates to the general domain of image coding.

The invention relates to a method for multi-view coding of a

of images and a corresponding method for decoding.

2. Prior art

Multi-view coding comprises the coding of sequences of images representing a same scene captured with different viewing angles. More generally, multi- view coding comprises the coding of sequences of images representing a same scene captured in different conditions, for example viewing angle, exposure, etc. Each sequence of images thus represents a particular view of the scene.

It is known in the art to code such views to use the redundant information that exists between them. For this purpose, one of the views is generally coded independently of the other views according to a known mono-view coding method. A coding method in accordance with the H.264/AVC coding standard is an example of such a mono-view coding method. In figure 1 , the images of the view j are thus coded independently of the image of the view i, for example in accordance with the H.264/AVC standard or the MPEG-2 Video standard. The images of the view i are then coded particularly from the images of the view j. More precisely, a block X of the view i is coded either typically according to a mono-view coding method (for example in accordance with H.264/AVC or MPEG-2 Video) from blocks of the same view i previously coded and reconstructed (arrows in solid lines on figure 1 ), or according to an inter-view coding method (for example in accordance with the Appendix H - the extension multi-view known as MVC - of the standard ITU-T Rec. H.264 \ ISO/I EC 14496-10) from a block Z not necessarily co-located of the image of the view j that coincides temporally with the image T of the view i to which belongs the block X (arrow in dotted line on figure 1 ). According to a variant known to the prior art shown in figure 2, the block X of the view i is coded according to an inter-layer coding method known under the coding name of scalable from the block X co-located and possibly neighbouring blocks (e.g. A, B, C, etc.) of the image T of the view j that coincides temporally with the image T of the view i to which belongs the block X (arrow in dots on figure 2). A coding method in accordance with the SVC (Scalable Video Coding) coding standard is an example of such an inter-layer coding method. Such an SVC type coding method comprises the prediction of the block X of the view i according to one of the prediction modes defined in Appendix G of the standard ITU-T REC. H.264 \ ISO/IEC 14496-10. The prediction of the block X comprises the prediction of data associated with the block X from corresponding data associated with the block X of the view j. These data are for example texture data, motion data (for example motion vectors, reference image indexes, partitioning into sub-blocks) or again residual data.

The SVC and MVC coding methods described remain limited in terms of coding efficiency and can not be used together.

3. Summary of the invention

The purpose of the invention is to overcome at least one of the disadvantages of the prior art.

The invention relates to a multi-view coding method of a sequence of images presented in the form of a first view and at least one additional view, each of the views comprising images of the sequence divided into blocks. The coding method comprises the following steps for:

- coding the images of the first view into a stream of coded data, and

- coding, in the stream of coded data, the images of the additional view from prediction data.

The coding of at least one current block of an image of the additional view known as the current image, comprises the following steps for:

- determining, according to a scalable SVC prediction mode, prediction data for the at least one current block from corresponding data associated with at least one reference block of an image of the first view, known as the reference image, non co-located,

- coding the at least one current block from prediction data, - coding for the current block data indicating the position of at least one reference block with respect to the block of the reference image co-located to the current block.

The method for coding according to the invention advantageously enables the coding efficiency of several views of the same scene to be augmented.

According to a particular aspect of the invention, the current image and the reference image do not coincide temporally and, for the at least one current block, an identification data of the reference image is coded in the stream of coded data.

According to a particular characteristic, the prediction data for the at least one current block are motion data and/or texture data and/or residual data.

According to another particular characteristic of the invention, the additional view comprises the same images as the images of the first view with a different exposure and, for the at least one current block, at least one item of data representative of an exposure variation between the current block and the reference block coded in the stream of coded data.

The invention also relates to a method for decoding a stream of coded data representative of a sequence of images presented in the form of a first view and at least one additional view, each of the views comprising images of the sequence divided into blocks, the method for decoding comprising the following steps for:

- reconstructing the images of the first view from the decoding of at least one first part of the stream of coded data, and

- reconstructing the images of the additional view from the decoding of at least one second part of the stream of coded data, and prediction data.

The reconstruction of at least one current block of an image of the additional view known as the current image, comprises the following steps for:

- decoding for the current block data indicating the position of at least one reference block of an image of the first view, known as the reference image, with respect to the block of the reference image, co-located to the current block, - determining, according to the SVC scalable prediction mode, prediction data for the at least one current block from corresponding data associated with at least one reference block, and

- reconstructing the at least one current block from prediction data.

4. List of figures

The invention will be better understood and illustrated by means of embodiments and advantageous implementations, by no means limiting, with reference to the figures in the appendix, wherein:

- figures 1 and 2 show two views of a same scene and the coding modes according to the prior art,

- figure 3 shows the coding method according to the invention,

- figures 4 and 5 show two views of a same scene and the coding modes according to the invention, and

- figure 6 shows the coding method according to the invention.

5. Detailed description of the invention

The invention relates to a multi-view coding method, for example in accordance with MVC, shown in figures 3 and 4.

In step 10, the images of a first view (view j) are coded typically according to a mono-view coding method. For example, a block X of the first view is coded by temporal prediction (INTER mode) or by spatial prediction (INTRA mode) from blocks of the same view previously coded and reconstructed. The residual data obtained by extracting, generally by subtracting, from the block X prediction data transformed, for example with a DCT (Discrete Cosine Transform), quantized and coded by VLC (Variable Length Coding) type entropy coding into a stream of coded data.

At step 12, prediction data are determined, according to an SVC type scalable prediction mode, for the block X of the image T of the view i, called the additional view, from corresponding data associated with at least one block K non co-located to an image T of the first view j. The block X is called the current block and the block K is called the reference block. As described in reference to figure 2, the prediction data are for example texture data, motion data or residual data.

In step 14, the block X of the image T of the view i is coded from data predicted in step 12.

In step 16, additional data (dx, dy) indicating the position of the reference block K of the first view with respect to the block co-located to the current block X are also coded. The block of the first view (view j) co-located to the block X of the additional view (view i) is also noted as X in figure 4. According to a variant shown in figure 5, the block K belongs to an image previously coded and reconstructed (e.g. image T-1 or image T+1 ) that does not coincide temporally with the image to which the block X of the view i belongs. In this case, the method for coding according to the invention comprises the coding for the block X of the view i of an item of identification data of the image to which block K belongs.

According to a particular characteristic of the invention, the images of the additional view are the same as the images of the first view with a different exposure, i.e. higher or lower. In this case, the method for coding according to the invention comprises the coding for the block X of the view i, of at least one item of data representative of the exposure variation between the block X of the view i and the block K. For example, two items of data representative of the exposure difference are coded: a weight value a and a threshold value β. For example, the intra texture T(X,i) or respectively the residue R(X, i) of the block X of the view i can be predicted from the intra texture T(K, j) or respectively from the residue R(K, j) of the block K of the view j according to the formula T(K, j) = a ^* T(X, i) + β, respectively R(K, j) = a ^* R(X, i) + β.

According to another particular characteristic, two additional views are coded, one of the two additional views comprising the same images as the images of the first view with a lower exposure and the other additional view comprising the same images as the images of the first view with a higher exposure.

The invention also relates to a method for decoding shown in figure 6. In step 20, the images of a first view (view j) are reconstructed typically according to a mono-view decoding method. For example, a block X of the first view is reconstructed by temporal prediction (INTER mode) or by spatial prediction (INTRA mode) from blocks of the same view previously reconstructed. This step implements inverse steps to those described for step 10 notably entropy decoding, inverse quantization, inverse transformation. In step 22, additional data (dx, dy) indicating the position of the reference block K of the first view with respect to the block co-located to the current block X are also coded. The block of the first view (view j) co-located to the block X of the additional view (view i) is also noted as X in figure 4.

At step 24, the prediction data are determined, according to a SVC type scalable prediction mode, for the block X of the image T of the view i from corresponding data associated with at least one block K non co-located of the image T of the first view j. As described in reference to figure 2, the prediction data are for example texture data, motion data or residual data.

In step 26, the block X of the image T of the view i is reconstructed from data predicted in step 12.