EP2392141A1

EP2392141A1 - Method and device for encoding an image using a prediction mask decoding method and device and corresponding signals and computer programs

Info

Publication number: EP2392141A1
Application number: EP10707587A
Authority: EP
Inventors: Isabelle Amonou; Matthieu Moinard; Nathalie Cammas
Original assignee: France Telecom SA
Current assignee: Orange SA
Priority date: 2009-01-28
Filing date: 2010-01-26
Publication date: 2011-12-07
Also published as: US9055294B2; CN102301713A; CN102301713B; US20110286521A1; WO2010086544A1

Abstract

The invention relates to a method for encoding at least one image, divided into macroblocks, one macroblock comprising a set of blocks de pixels. According to eh invention, such a method comprises the following steps for at least one current macroblock of a current image, at least two iterations of the following steps: allocation (31) of a priority level for encoding to at least one pixel of the current macroblock, selection (32) of a pixel with the highest priority level called the priority pixel, prediction (33) of a set of pixels including said priority pixel and an encoding step (35) for a remainder from prediction for a region made up of predicted pixels of said macroblock, called the predicted region, said encoding step being used as and when said predicted region meets a predetermined encoding criterion.

Description

Method and apparatus for coding an image using a prediction mask, decoding method and device, signal and corresponding computer programs. FIELD OF THE INVENTION The field of the invention is that of encoding and decoding of images, and in particular of a video stream consisting of a series of successive images. More specifically, the invention applies to the compression of images or image sequences using block transforms.

The invention can notably apply to video coding implemented in current (MPEG, H.264, etc.) or future (ITU-T / VCEG (H.265) or ISO / MPEG (HVC) video encoders) . 2. Prior Art

Many techniques of video data compression are already known. Among these, many video coding techniques use a block representation of the video sequence, such as those implementing video compression standards from the MPEG organization (MPEG-I, MPEG-2, MPEG-1 4 part 2, ...) or ITTU-T (H.261, ..., H.264 / AVC). Thus, according to the H.264 technique, each image can be sliced (in English "slice"), themselves cut into macroblocks, which are then subdivided into blocks. A block consists of a set of pixels. According to the H.264 standard, a macroblock is a square block of size equal to 16x16 pixels, which can be split into blocks of 8x8, 16x8 or 8x16 size, the 8x8 blocks can then be cut into sub-blocks of size 4x4, 8x4 or 4x8.

According to the known techniques, the macroblocks or the blocks can be coded by intra-image prediction or inter-image prediction. In other words, a macroblock or block may be encoded by: a temporal prediction, that is, with reference to a reference block or macroblock belonging to one or more other images; and / or - a spatial prediction, depending on the blocks or macroblocks of the current image.

In the case of a spatial prediction, the prediction of a current block can only be made from blocks that have been previously coded, using a directional extrapolation technique of the coded texture values. decoded on the neighboring blocks. These blocks are said to belong to the "causal neighborhood" of the current block, comprising blocks located before the current block in a predetermined direction of travel of the blocks in the image.

Thus, to predict a macroblock or a block from its causal neighborhood, nine intra prediction modes are used according to the ITU-T H.264 standard. These prediction modes comprise eight modes corresponding to a given orientation for copying the pixels from the previously coded-decoded neighboring blocks (vertical, horizontal, diagonal down left, diagonal down right, vertical right, vertical left, horizontal amount, horizontal descending), and a mode corresponding to the average of the pixels adjacent to the block from the neighboring blocks.

Unfortunately, spatial prediction alone is insufficient, hence the need to code a prediction error. Thus, for each block is coded a residual block, also called prediction residue, corresponding to the original block decreased by a prediction. For that, the coefficients of this block are quantized, after a possible transformation (for example the DCT transform - in English "Discrète cosine transform", in French "transformed in discrete cosine"), then coded by an entropic coder. At the encoder, the prediction mode chosen is that which makes it possible to obtain the best rate-distortion compromise.

Thus, in intra coding mode, a prediction of the texture values of the current block is established from the coded-decoded texture values of the neighboring blocks, and then a prediction residue is added to this prediction.

In the case of time prediction, the ITU-T H.264 standard uses motion-displacement to predict a block or macroblock from its time neighborhood. The motion vector is then encoded and transmitted. Alternative methods of intra prediction have recently been proposed, based in particular on the correlations between neighboring pixels.

For example, the article "Intra prediction by template matching" of Tan et al. presents such a texture synthesis technique based on an intra prediction, also called "template matching". A simplified diagram of this technique is illustrated in FIG.

This technique makes it possible to synthesize a pixel p (or a group of pixels) in a target area C of the image, taking into account a source area S of the same image. It is based on correlations between neighboring pixels. Thus, the value of each pixel p of the target area C is determined by comparing (11) the pixels N (p) belonging to the causal neighborhood of the pixel p, defining a "template" or "mask" of the pixel p, with all the neighbors of the source area S. The mask N (p) is made up of previously coded / decoded pixels, which avoids the propagation of errors. If we find a region of the source zone S similar to the mask defined by N (p), then we assign (12) to the pixel p of the target zone C, the pixel (or group of pixels) of the source zone S presenting the most similar neighborhood.

Conventionally, the region closest to the mask in the source image, corresponding to a region similar to the mask, is chosen on criteria for minimizing the quadratic or absolute error according to the following formula: N (q ') = aτg min d (N (p), N (q))

N (q) ≡S with:

q a pixel of the source zone S;

a function measuring the distance between two masks according to the minimization criterion used.

According to this technique, the pixels are synthesized in a fixed and predetermined order, generally from top to bottom and from left to right (in English "raster scan order").

The classic "template matching" technique makes it possible to synthesize textures made up of random structures. On the other hand, it is not very effective when applied to natural images, composed of mixed textures and contours.

The "template matching" technique has also been extended to the block coding of an H.264 encoder by Wang et al., As described in the document "Priority-based template matching intra prediction. It is thus possible to code / decode a macroblock using blocks (composed of 4x4 pixels according to the H.264 standard) reconstructed by the "template matching" technique.

As illustrated in FIG. 2, in the application by Wang et al., The blocks are conventionally processed in a left-to-right and top-to-bottom ("raster scan") order of the blocks in the macroblock. Therefore, the macroblock is encoded / decoded (or rebuilt) block by block. Processing the blocks one after the other makes it possible to use the pixels of a previously coded / decoded block as "source zone" for a current block, according to the order of travel.

Unfortunately, this technique is not optimal because it forces the reconstruction of the macroblock to follow blocks of 4x4 pixels.

There is therefore a need for new image coding / decoding techniques implementing a "template matching" type of synthesis technique, making it possible to improve these techniques of the prior art.

3. DISCLOSURE OF THE INVENTION The invention proposes a new solution which does not have all of these disadvantages of the prior art, in the form of a method of encoding at least one image, divided into zones of standard size, called macroblocks, a macroblock comprising a set of blocks of pixels.

According to the invention, such a method implements, for at least one current macroblock of a current image: at least two iterations of the following steps:

assigning a coding priority level to at least one pixel of the current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion; - selection of a pixel with the highest priority level, says priority pixel;

Prediction of a set of pixels comprising the priority pixel, said target set, implementing the following substeps:

"identifying a mask corresponding to a predetermined template region, the mask and the target set forming a block;

Searching in the previously coded blocks of the current image or of a reference image of a source block comprising a zone similar to the mask; "constructing the target set from the source block, delivering predicted pixels; and a step of encoding a prediction residue for a region of predicted pixels of the macroblock, said predicted region, by determining a difference between the predicted pixels of the predicted region and the original pixels belonging to the current (uncoded) image.

According to the invention, the coding step is implemented as soon as the predicted region complies with a predetermined coding criterion.

The invention thus proposes to improve the existing "template matching" techniques by working directly on a macroblock, and by adapting the order of the macroblock pixels for the prediction of the macroblock.

This optimizes the quality of the prediction and thus the quality of the coding and decoding of images, while reducing the bit rate necessary for coding.

It will be recalled that, according to the skilled person, the efficiency of a template matching technique applied to an entire macroblock is limited because the further one moves towards the end of the macroblock, the more predicted pixels used for the prediction are away from the pixels to predict. Indeed, the pixels predicted at the beginning of the macroblock have not yet been coded / decoded (the prediction error has not yet been coded), the predicted pixels do not provide a sufficiently good prediction, and may propagate errors. errors towards the end of the macroblock and to increase by the same amount the cost of coding the residue. It is thus possible to signal once, for the entire macroblock, that a template matching technique is used, without having to report it for each block.

In addition, whereas according to the conventional "template matching" technique, the pixels are synthesized in a fixed and predetermined order, it is proposed according to the invention to adapt this order of travel as a function of a priority level assigned to the pixels. and recalculate this course order when a target pixel array is predicted.

The direction of travel of the pixels inside a macroblock is therefore adaptive, whereas according to the techniques of the prior art, this order of travel was predefined and fixed.

Finally, it is noted that the prediction residue, corresponding to the difference between the predicted pixels of the predicted region and the original pixels belonging to the current image, can be practically nil. This means that the prediction is sufficient (for example for a targeted application: low speed, low quality, etc.). In this case, the obtained prediction residue is small and can be canceled during the quantization of the residue.

For example, the coding criterion belongs to the group comprising: a predicted number of pixels greater than a predetermined threshold; a predetermined configuration of the predicted pixels.

For example, such a configuration is in agreement with the transforms used by the encoder, and corresponds to a 4x4 block in the case of an H.264-AVC encoder. Such a configuration can also be processed by a transform implemented in the encoder / decoder, and corresponds to a 2x8, or 8x2, or 16x1, or 1x16 block, or any region of 16 pixels in the case of a transform. SA-DCT type (in English "Shape Adaptive DCT", in French "Adaptive form discrete cosine transform").

Thus, as soon as enough pixels are predicted (or restored), it is possible to code and decode the corresponding predicted region. Pixels thus encoded / decoded (or reconstructed) can then be used for the prediction of following regions of the macroblock It is considered that these coded / decoded pixels now belong to a "source" zone as defined in relation with the prior art

For example, the predicted region comprises sixteen predicted pixels. In other words, as soon as sixteen pixels are predicted (forming a region of 4x4 pixels, 2x8 pixels, 8x2 pixels, 16x1 pixels or 1x16 pixels), the residue associated with this region, and we code / decode this region

According to a particular embodiment of the invention, the priority determination criterion takes account of at least one parameter belonging to the group comprising.

A number of previously predicted or coded pixels, said available pixels, adjacent to said at least one pixel,

a presence of a contour which follows a contour coming from a set of previously predicted or coded pixels, of which at least one pixel of said set is adjacent to said at least one pixel,

a number of previously predicted or coded pixels belonging to said mask,

A statistical parameter representative of an activity of the vicinity of said pixel, a number of candidates similar to said mask, defined for example by the sum of the absolute differences, as described in relation with the prior art

Thus, the direction of travel of the pixels inside a macroblock takes into account parameters making it possible to optimize the coding by improving the efficiency of the prediction and therefore the quality of the reconstruction of the pixels.

For example, one of the parameters taken into account to assign a priority to a pixel corresponds to the number of available pixels already coded-decoded (or reconstructed) that can be used for coding this pixel.

This parameter can have for example a value corresponding directly to the number of pixels available Another parameter takes into account the presence of an incident contour in a pixel, that is to say a contour that is also present in an already coded-decoded adjacent pixel.

Thus, if a diagonal contour passes through a macroblock from an already coded-decoded macroblock, the technique according to one embodiment of the invention adapts the order of travel according to the direction of the contour, and the order of travel. obtained diagonal improves the quality of the prediction and thus the speed and quality of the decoding. This parameter indicating the presence of an incident contour in a pixel (the pixel belonging to the contour), also noted as a contour parameter, may have for example a value of 1.

Yet another parameter takes into account the number of previously predicted or coded pixels present in the mask defined near the pixel. This gives preference to the prediction of a target set for which the corresponding mask comprises the largest number of pixels already predicted. Then, the values of the parameters described above can be added, or multiplied, to obtain a priority level assigned to each pixel adjacent to a previously predicted pixel.

According to another aspect of the invention, at least one of the iterations is a refinement iteration, in which the assignment step assigns a coding priority level to at least one pixel of the current macroblock adjacent to at least one previously coded pixel.

Here coded / decoded pixels are considered, and not only predicted. Indeed, coded / decoded pixels can be given greater confidence than only predicted pixels. It is recalled to this effect that if the predicted pixels do not provide a sufficiently good prediction, they may propagate errors towards the end of the macroblock and increase the coding cost of the residue. Pixels already coded / decoded are therefore used to refine the prediction.

The invention also relates to a computer program comprising instructions for implementing the encoding method described above when the program is executed by a processor. Thus, the coding method according to the invention can be implemented in various ways, in particular in hard-wired form or in software form.

In another embodiment, the invention relates to a device for encoding at least one image, divided into standard size areas, called macroblocks, a macroblock comprising a set of blocks of pixels.

According to the invention, such a device comprises the following means, activated for at least one current macroblock of a current image and for at least two iterations:

means for assigning a coding priority level to at least one pixel of the current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion;

- Means for selecting a pixel having the highest priority level, said priority pixel; means for predicting a set of pixels comprising the priority pixel, said target set, comprising:

means for identifying a mask corresponding to a predetermined model region, the mask and the target assembly forming a block; search means in the previously coded blocks of the current image or of a reference image a source block comprising a zone similar to the mask;

means for constructing the target set from the source block, delivering predicted pixels, and coding means for a prediction residue for a predicted region of predicted pixels of the macroblock, said predicted region, by determination of a difference between the predicted pixels of the predicted region and the original pixels belonging to the current image (uncoded), the coding means being activated as soon as the predicted region complies with a predetermined coding criterion. Such a coding device is particularly suitable for implementing the coding method described above. This is, for example, a video coder of the MPEG or H.264 type, or according to a future video compression standard, for example H.265. The invention also relates to a signal representative of at least one image coded according to the coding method described above.

Such a signal comprises at least one piece of information representative of the coding criterion, such as a threshold corresponding to the minimum number of pixels that must be predicted to reconstruct a predicted region. This signal may of course include the various characteristics relating to the coding method according to the invention.

The invention also relates to a method for decoding a signal representative of at least one image, divided into zones of standard size, called macroblocks, a macroblock comprising a set of blocks of pixels. According to the invention, such a method implements at least two iterations of the following steps, for at least one current macroblock of a current image:

assigning a coding priority level to at least one pixel of the current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion; selecting a pixel having the highest priority level, said priority pixel;

"search in the previously coded blocks of the current image or reference image of a source block including a mask-like area;" construction of the target set from the source block, delivering predicted pixels; and in that said method also comprises:

a step of receiving at least one prediction residue;

a step of reconstructing a region formed of predicted pixels of the macroblock, said predicted region, by adding a prediction residue corresponding to the predicted region, the reconstruction step being implemented as soon as the predicted region complies with a predicted region; predetermined coding criterion.

The features and advantages relating to this decoding method are the same as those of the coding method. Therefore, they are not detailed further.

Such a decoding method is able to decode a signal as described above.

Thus, according to a particular characteristic, such a method comprises a step of receiving at least one piece of information representative of the coding criterion.

The invention also relates to a computer program comprising instructions for implementing the decoding method described above, when the program is executed by a processor. Thus, the decoding method according to the invention can be implemented in various ways, in particular in hard-wired form or in software form.

In another embodiment, the invention relates to a device for decoding a signal representative of at least one image, divided into zones of standard size, called macroblocks, a macroblock comprising a set of blocks of pixels.

means for assigning a coding priority level to at least one pixel of the current macroblock, adjacent to at least one pixel previously predicted, according to a predetermined priority determination criterion;

"means for constructing the target set from the source block, delivering predicted pixels, as well as:

means for receiving at least one prediction residue;

means for reconstructing a region formed of predicted pixels of the macroblock, said predicted region, by adding a prediction residue corresponding to the predicted region, activated as soon as the predicted region complies with a predetermined coding criterion.

Such a decoding device is particularly suitable for implementing the decoding method described above. This is for example a MPEG or H.264 video decoder, or according to a future video compression standard, for example H.265.

4. List of figures

Other features and advantages of the invention will appear more clearly on reading the following description of a particular embodiment, given as a simple illustrative and nonlimiting example, and the appended drawings, among which: FIG. 1, commented on in relation to the prior art, presents a simplified diagram of the so-called "template matching"technique; FIG. 2 illustrates an example of direction of travel for the coding of the blocks of a macroblock; - Figure 3 shows the main steps implemented in the coding according to a particular embodiment of the invention; FIGS. 4A to 4G illustrate the application of the coding method of FIG. 3 to a macroblock; Figure 5 shows the main steps implemented in the decoding according to a particular embodiment of the invention; Figures 6 and 7 respectively show the structure of a coding device and a decoding device according to a particular embodiment of the invention.

5. Description of an embodiment of the invention 5.7 General principle

The general principle of the invention is based on the adaptive coding / decoding of a macroblock, making it possible to predict pixels, and then to code / decode predicted pixel regions, as soon as there is a sufficient number of predicted pixels for each pixel. the transformation and coding of the residue. Thus, the coding / decoding of the macroblock does not require the coding / decoding of the different blocks of the macroblock, one after the other, according to a predefined order of travel.

FIG. 3 illustrates the main steps implemented on the coding side, for at least one current macroblock of a current image. The line dividing the previously predicted pixels from the not yet predicted pixels is called the "prediction boundary", and the "coding boundary" is a line separating the already coded / decoded pixels from the pixels not yet coded / decoded.

Thus, the coding method implements at least two iterations of the following steps: assigning a coding priority level to at least one pixel of the current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion. In other words, a priority level is assigned to the not yet predicted pixels located at the prediction boundary. During the first coding iteration of a macroblock, the prediction and coding boundaries are superimposed;

selecting 32 a pixel having the highest priority level, said priority pixel p, from the pixels located at the prediction boundary; - Prediction 33 of a set of pixels comprising the priority pixel, said target set P, implementing the following substeps:

"identification 331 of a mask Tp corresponding to a predetermined model region, the mask Tp and the target set P forming a block;" search 332 in the previously coded blocks of the current image or of a reference image of a source block comprising a zone similar to the mask Tp; Construction 333 of the target set P from the source block, delivering predicted pixels. The prediction boundary is thus displaced to encompass the new predicted pixels.

It is then tested (34) whether a region formed of predicted pixels of the macroblock, said predicted region, meets a predetermined coding criterion.

If this is not the case (341), then the previous steps are repeated by returning to step 31 of assigning a coding priority level.

If the predicted region respects (342) the predetermined coding criterion, then the coding method implements a step of coding a prediction residue for the predicted region of the macroblock, by determining a difference between the predicted pixels. of the predicted region and original pixels belonging to the current image. It is recalled that the prediction residue can be virtually zero. In this case, the prediction residue obtained is canceled during the quantization step.

At the end of the coding step, it is considered that the predicted region is coded / decoded. As a result, the coding boundary is shifted to encompass the new coded / decoded pixels. These new coded / decoded pixels are now considered as pixels belonging to the source zone of the image, which can be used for the prediction of unpredicted pixels, or the refinement of previously predicted pixels.

5.2 Description of a particular embodiment for the implementation of the coding

FIGS. 4A to 4G illustrate the application of a particular embodiment of the coding method to the case of intra prediction in an H.264 coder for example, using macroblocks of size 16 × 16. cut into blocks of 4x4 size. Of course, this is just an example. Indeed, the invention also applies to an image encoder using regions of arbitrary size and shape, in the context of intra or interimage prediction.

According to this application example, the current macroblock MB is composed of sixteen blocks each comprising 4x4 pixels. This macroblock MB is adjacent, in the image, to three already encoded-decoded macroblocks (located above the current macroblock - MBj, to the left of the current macroblock - MBL, ^and above to the left and the current macroblock - MBJL). . By "adjacent" is meant here a macroblock that touches the boundary of the current macroblock.

According to this embodiment, it is sought to restore the current MB macroblock optimally, respecting an order of priority for coding / decoding (or reconstruction) information.

Macroblocks MBj, MBL ^and MBJL- previously coded / decoded (represented by diagonal hatching in both directions) are considered to belong to a source area of the image, which can be used for the prediction of regions not yet predicted. FIGS. 4B to 4G each illustrate the result obtained after a prediction iteration as described with reference to FIG. 3, for the coding of the current macroblock MB.

As illustrated in FIG. 4B, the previously coded / decoded macroblocks MBj, MBL ^and MBJL define, with the current macroblock MB, a coding boundary FR. This FR coding boundary (illustrated in the other figures 4C to 4G in short dotted lines) is superimposed with a prediction boundary Fp (illustrated in long dashed lines) during the first iteration.

During a first iteration according to the algorithm described in FIG. 3, a coding priority level is assigned to the unpredicted pixels of the macroblock MB located at the prediction boundary Fp. For example, a priority level is assigned to all the pixels of the MB macroblock adjacent to a previously predicted pixel MBJ, MBL and MBJL macroblocks previously coded / decoded. Consider for example that the pixel p, located in the top left corner of macroblock MB, has the highest priority level. Indeed, this pixel is adjacent to five pixels previously coded / decoded (and therefore predicted). This pixel is called "priority pixel". One then seeks to predict a target set P, represented by horizontal hatching in FIG. 4B, comprising the priority pixel p. This target set corresponds for example to the first three lines and first three columns of the first block MB macroblock.

To do this, a mask Tp is identified, such that the mask Tp and the target set P form a block (see FIG. 4A for the sake of clarity).

In the previously coded blocks of the current image (in the case of an intra prediction) or of a reference image (in the case of an inter prediction), a source block comprising a similar zone Tn in the mask is then searched. Tp. This search region comprising a set of previously coded / decoded (or reconstructed) blocks, also called source zone, is delimited by the coding boundary. In other words, we look in the source area for a "template" Tn (or mask) closest to the mask Tp, and then we build the target set P from the source block thus identified. To do this, it is possible to copy in the target set P the pixels of the source block identified that do not belong to the similar zone Tn. This "template matching" technique is more precisely described in relation to the prior art. In this way, a prediction of the target set P, delivering predicted pixels, is obtained.

Since the number of predicted pixels is not sufficient to code / decode a region of the macroblock MB, the preceding steps are repeated during a second iteration. As illustrated in FIG. 4C, it can be seen that the prediction boundary Fp has moved to encompass the new predicted pixels. In contrast, the FR encoding boundary has not changed.

More precisely, during a second iteration, a new coding priority level is assigned to the unpredicted pixels of the MB macroblock adjacent to a previously predicted pixel. These non-predicted pixels correspond to the pixels located at the prediction boundary Fp.

For example, it is considered that the pixel p, corresponding to the pixel located on the fourth line, fourth column of the macroblock MB, has the highest priority level. For example, this priority pixel p is located on a contour crossing the target set defined at the first iteration.

It is sought to predict a new target set P, represented by horizontal hatching in FIG. 4C, comprising the priority pixel p. To do this, we identify a new mask Tp, such as the mask Tp and the target set

P form a block. Note that the shape of the target set P and the mask Tp (that is to say the number of pixels and the configuration of the pixels) does not change during the different iterations. On the other hand, the pixels present in the target set P and in the mask Tp differ from one iteration to the other.

A source block comprising a similar zone Tn to the mask Tp is then searched in the source zone. This research area is the same as for the first iteration, since the FR coding boundary has not evolved.

In this way, a prediction of the target set P, delivering predicted pixels, is obtained.

Since the number of predicted pixels is not sufficient to code / decode a region of the macroblock MB, the preceding steps are repeated during a third iteration (the result of which is illustrated in FIG. 4D) and then a fourth iteration. (whose result is illustrated in FIG. 4E), then a kth iteration (the result of which is illustrated in FIG. 4F).

At the conclusion of the third and fourth iterations, the number of predicted pixels is not sufficient to code / decode a region of the macroblock MB. As a result, the FR encoding boundary has not changed. In contrast, the prediction boundary Fp has shifted to encompass the new predicted pixels.

At the end of the kth iteration, it is considered that the number of predicted pixels is sufficient to code / decode (or reconstruct) a predicted region of the macroblock MB, by coding a prediction residue associated with this region. For example, a sufficient number of pixels has been predicted for these predicted pixels to be used for transformation and coding of the residue. For example, the first block B 1 (located at the top left of the macroblock MB) is totally predicted. Of course, this residue coding step can be implemented to code / decode (or reconstruct) a predicted region of different size, such as a region consisting of 2x2 predicted pixels, 4x4 predicted pixels, 8x8 predicted pixels, 2x8 predicted pixels, 8x2 predicted pixels, 16x1 predicted pixels, 1x16 predicted pixels, etc. One could therefore reconstruct a block of data as soon as 16 pixels have been predicted (for example 16 pixels of the same line, or of the same column, ...).

In other words, this step of coding the residue is implemented as soon as the predicted region respects a predetermined coding criterion, such as a predicted number of pixels greater than a predetermined threshold, or a specific configuration of the predicted pixels. At the end of this coding step, the first block B1 of the macroblock MB is reconstructed (that is to say encoded and decoded with its residue). Therefore, the FR encoding boundary is shifted to encompass this new encoded / decoded B block, which is now part of the source area of the image, and can serve as a reliable source for predicting the remaining regions to be restored from the macroblock. MB. It is recalled that it is however not possible to use, for the prediction, predicted but not reconstructed pixels.

During an iteration k + 1, the result of which is illustrated in FIG. 4G, it is therefore possible to predict the target set P using the block B1. In addition, it is also possible to refine the prediction of certain pixels when a region has been coded / decoded.

For example, as illustrated in FIG. 4G, it is possible to refine the prediction of the predicted region located between the prediction Fp and FR coding boundaries, taking into account the new source zone. Thus, the invention according to this embodiment proposes to code / decode

(or rebuild) a region of the macroblock as soon as enough pixels have been predicted and can be used for residue transformation and coding. These pixels are then coded (predicted pixels to which a prediction residue is added) and can in turn be used for prediction. Thus the invention proposes several advantages resulting from the implementation of at least one embodiment:

the algorithm can progress freely in the macroblock by restoring / predicting image regions "on horseback" over several blocks; the reconstruction of the blocks (4x4 size according to the H.264 standard) is not imposed according to a predetermined order.

We also note that the proposed solution can be integrated without difficulty in a video coder based macroblock type H.264, without substantially changing its operation. For that, it is enough for example to add a new mode of spatial prediction of the macroblocks, or to replace several existing modes little used in the context of the present invention. In particular, the rate-distortion optimization process remains unchanged.

5.3 Signal

Once the image or the coded images, the signal representative of the encoded image (s) according to the coding method described above can be transmitted to a decoder and / or stored on a recording medium.

This signal can carry particular information. For example, this signal may carry at least one piece of information representative of the coding criterion, such as a threshold corresponding to the minimum number of pixels that must be predicted to reconstruct a predicted region.

This signal can also transmit information representative of the priority determination criterion. Indeed, the priority of the pixels to be predicted, located at the level of the prediction boundary, can be calculated at the encoder and transmitted to the decoder, or alternatively recalculated to the decoder. This signal can also transport the prediction residue (s) obtained by comparing the current image and the predicted image.

5.4 Decoding

The main decoding steps of a signal representative of at least one coded image as described previously, implemented in a video decoder of the H.264 type, for example, are now presented in relation with FIG.

Such a decoder implements the following steps, for at least one current image:

at least two iterations It 51 of the following steps, for at least one current macroblock of a current image: assignment of a coding priority level to at least one pixel of the current macroblock, adjacent to at least one pixel previously predicted according to a predetermined priority determination criterion; selecting a pixel having the highest priority level, said priority pixel; o prediction of a set of pixels comprising said priority pixel, said target set;

a step of receiving 52 of at least one prediction residue;

a step of reconstruction 53 of a region formed of predicted pixels of the macroblock, said predicted region, by adding a prediction residue corresponding to the predicted region, the reconstruction step being implemented as soon as said predicted region respects a predetermined coding criterion.

At least one piece of information representative of this coding criterion may be conveyed by the signal. In this case, the decoding method comprises an additional step of receiving the information representative of this coding criterion. Similarly, the decoding method may implement an additional step of receiving information representative of the priority determination criterion. The iterations 51 are similar to those described in the coding, and are not described in more detail here.

In a simplified manner, the decoding algorithm repeats the following steps until the macroblock is completely reconstructed:

determining a priority level for pixels adjacent to already predicted pixels (located at the prediction boundary);

prediction of a target set associated with the highest priority pixel, using a "template matching" technique;

as soon as a sufficient number of new pixels are restored, adding a prediction residue associated with the region thus predicted, making it possible to reconstruct this region. The pixels of the reconstructed region then become source pixels (belonging to a source area) that can be used for the prediction of the subsequent regions of the macroblock. 5.5 Encoder and decoder structures

Finally, with reference to FIGS. 6 and 7, the simplified structures of a coding device and a decoding device introducing respectively implement a coding technique and a decoding technique as described above.

A coding device as illustrated in FIG. 6 comprises a memory 61 comprising a buffer memory, a processing unit 62, equipped for example with a microprocessor μP, and driven by the computer program 63, implementing the method coding according to the invention.

At initialization, the code instructions of the computer program 63 are for example loaded into a RAM before being executed by the processor of the processing unit 62. The processing unit 62 receives a current image at code. The microprocessor of the processing unit 62 implements the steps of the coding method described above, according to the instructions of the computer program 63, for coding the current image. For this purpose, the coding device comprises, in addition to the buffer memory 61, means for assigning a coding priority level to at least one pixel of a current macroblock, means for selecting a pixel presenting the level of highest priority, means for predicting a set of pixels comprising the priority pixel and means for coding a prediction residue for a region formed of predicted pixels of the macroblock.

These means are controlled by the microprocessor of the processing unit 62. A decoding device as illustrated in FIG. 7 comprises a memory 71 comprising a buffer memory, a processing unit 72, equipped for example with a microprocessor μP, and controlled by the computer program 73, implementing the decoding method according to the invention.

At initialization, the code instructions of the computer program 73 are for example loaded into a RAM memory before being executed by the processor of the processing unit 72. The processing unit 72 receives as input a signal representative of at least one coded picture. The microprocessor of the processing unit 72 implements the steps of the decoding method described above, according to the instructions of the computer program 73, to determine reconstructing the current image. For this, the decoding device comprises, in addition to the buffer memory 71, means for assigning a coding priority level to at least one pixel of a current macroblock, means for selecting a pixel having the highest priority level, means for predicting a set of pixels comprising the priority pixel, means for receiving at least one prediction residue and means for reconstructing a region formed of predicted pixels of the macroblock.

These means are controlled by the microprocessor of the processing unit 72.

Claims

A method for encoding at least one image, divided into macroblocks, a macroblock comprising a set of blocks of pixels, characterized in that said method implements, for at least one current macroblock of a current image: at least one two iterations of the following steps:

assigning (31) a coding priority level to at least one pixel of said current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion; selecting (32) a pixel having the highest priority level, said priority pixel;

prediction (33) of a set of pixels comprising said priority pixel, said target set, implementing the following substeps: "identification (331) of a mask corresponding to a predetermined model region adjacent to said target set, said mask and said target assembly forming a block;

"searching (332) in the previously encoded blocks of said current image or of a reference image of a source block including a region similar to said mask;

"constructing (333) said target set from said source block, delivering predicted pixels; and a step of encoding (35) a prediction residue for a region of predicted pixels of said macroblock, said predicted region, by determining a difference between the predicted pixels of said predicted region and corresponding original pixels, said original pixels belonging to said current image, said coding step being implemented as soon as said predicted region complies with a predetermined coding criterion, before all the pixels of said macroblock are predicted.

2. Coding method according to claim 1, characterized in that said encoding criterion belongs to the group comprising:

a number of predicted pixels greater than a predetermined threshold;

a predetermined configuration of the predicted pixels.

3. coding method according to claim 1, characterized in that said predicted region comprises sixteen predicted pixels.

4. Coding method according to claim 1, characterized in that said priority determination criterion takes account of at least one parameter belonging to the group comprising:

a number of previously predicted or coded pixels adjacent to said at least one pixel;

a presence of a contour which follows a contour coming from a set of previously predicted or coded pixels, of which at least one pixel of said set is adjacent to said at least one pixel;

a number of previously predicted or coded pixels belonging to said mask;

a statistical parameter representative of an activity of the neighborhood of said pixel;

a number of candidates similar to said mask.

The coding method as claimed in claim 1, wherein at least one of said iterations is a refinement iteration, during which said assignment step assigns a coding priority level to at least one pixel of said current macroblock. adjacent to at least one previously coded pixel.

Computer program comprising instructions for implementing the coding method according to claim 1 when said program is executed by a processor.

7. Device for coding at least one image, divided into standard size zones, called macroblocks, a macroblock comprising a set of blocks of pixels, characterized in that said device comprises the following means, activated for at least one current macroblock of a current image and for at least two iterations:

means for assigning (31) a coding priority level to at least one pixel of said current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion;

- Selection means (32) of a pixel having the highest priority level, said priority pixel;

means for predicting (33) a set of pixels comprising said priority pixel, said target set, comprising: means for identifying (331) a mask corresponding to a predetermined model region adjacent said target set, said set mask and said target assembly forming a block;

Search means (332) in the previously coded blocks of said current image or of a reference image of a source block comprising a zone similar to said mask;

• building means (333) of said target set from said source block, delivering predicted pixels; and coding means (35) of a prediction residue for a region formed of predicted pixels of said macroblock, said predicted region, by determining a difference between the predicted pixels of said predicted region and corresponding original pixels, said original pixels belonging to said current image, said encoding means being activated as soon as said predicted region respects a predetermined coding criterion before all the pixels of said macroblock are predicted.

8. Signal representative of at least one coded image according to the coding method of claim 1, characterized in that it comprises at least one piece of information representative of said coding criterion.

9. A method of decoding a signal representative of at least one image, divided into standard size zones, called macroblocks, a macroblock comprising a set of blocks of pixels, characterized in that said method implements at least two iterations. (51) next steps, for at least one current macroblock of a current image:

assigning a coding priority level to at least one pixel of said current macroblock, adjacent to at least one previously predicted pixel, according to a predetermined priority determination criterion;

selecting a pixel having the highest priority level, said priority pixel;

prediction of a set of pixels comprising said priority pixel, said target set, implementing the following substeps:

"identifying a mask corresponding to a predetermined template region adjacent to said target set, said mask and said target set forming a block;

Searching in the previously coded blocks of said current image or of a reference image of a source block comprising a zone similar to said mask;

Constructing said target set from said source block, delivering predicted pixels; and in that said method also comprises:

a step of receiving (52) at least one prediction residue; a step of reconstructing (53) a region formed of predicted pixels of said macroblock, said predicted region, by adding a prediction residue corresponding to said predicted region, said reconstruction step being implemented as soon as said predicted region respects a predetermined coding criterion before all the pixels of said macroblock are predicted.

10. Decoding method according to claim 9, characterized in that it comprises a step of receiving at least one piece of information representative of said coding criterion.

11. Computer program comprising instructions for implementing the decoding method according to claim 9 when said program is executed by a processor.

12. A device for decoding a signal representative of at least one image, divided into standard size zones, called macroblocks, a macroblock comprising a set of blocks of pixels, characterized in that said device comprises the following means (51) , activated for at least one current macroblock of a current image and for at least two iterations: means for assigning a coding priority level to at least one pixel of said current macroblock, adjacent to at least one pixel previously predicts, according to a predetermined priority determination criterion;

- Means for selecting a pixel having the highest priority level, said priority pixel;

means for predicting a set of pixels comprising said priority pixel, said target set, comprising:

means for identifying a mask corresponding to a predetermined template region adjacent to said target set, said mask and said target set forming a block;

search means in the previously coded blocks of said current image or of a reference image of a source block comprising a zone similar to said mask; means for constructing said target set from said source block, delivering pixels; predicted; and in that said device also comprises:

means for receiving (52) at least one prediction residue; means for reconstructing (53) a region formed of predicted pixels of said macroblock, said predicted region, by adding a prediction residue corresponding to said predicted region, activated as soon as said predicted region respects a predetermined coding criterion before all the pixels of said macroblock are predicted.