FR3036016A1 - PREDICTIVE CODING / DECODING OF IMAGES HAVING ARTIFACT USING EXTRAPOLATION - Google Patents

Abstract

The invention relates to a prediction coding / decoding of a current block of at least one current image, in which a motion vector (mv) is obtained pointing to a reference block (brief) of a zone of pixels previously decoded, said pixel area comprising an artifact having a determined position. In particular: - determining (S3), depending on the position of the artifact in said pixel area, if the reference block comprises at least a part of the artifact, - if necessary, it corrects (S4) the reference block by correcting pixels at least of said part of the artifact by extrapolation from other pixels of said pixel area, and - using the corrected reference block (brief *) for decoding (S6) by prediction.

Description

The present invention relates to the coding / decoding, predictive, of a current image (fixed or of a sequence) having, in the same image, at least one artifact having at least one artifact. a determined position. This artifact can be for example a periodic pattern in the same image, as encountered especially in the so-called "integral" images for a three-dimensional (3D) display, such as 3D video images.

Indeed, current 3D video techniques have certain limitations. For example, the stereoscopy is based on the use of 3D glasses (informative glasses on each eye with for example a spatial phase difference allowing a 3D reconstruction with the images sent to each eye). However, a substantial disadvantage results from the use of this technology. For example, a time lag between the right and left lens images (including only one time-shift image only) can make viewing uncomfortable. An alternative technique to stereoscopy concerns video services called "3D without glass". In this type of technique, an image is reproduced a multiplicity of times at different points of view by placing a network of microlenses in front of an object of origin. For this object, we obtain a multiplicity of micro-images, all of which is called "integral image" (Figure 9). Each micro-image is different according to the angle of view and the position of the corresponding micro-lens. Thus, rather than displaying a 2D image, identical according to the observation directions, we obtain a reproduction of the light field on more than two dimensions (in four dimensions, more exactly). More particularly, with reference to FIG. 1, this "integral" imaging technique, also known as holoscopic or plenoptic imaging, consists in capturing a multiplicity of views by using an optical lens array placed between an object and a video camera. The network typically comprises a set of micro-lenses each of which LEN operates as a low-resolution individual micro-camera (since the number of pixels n × m allocated to each LEN micro-lens is in principle limited by the total number 3036016 2 pixels of MAT matrix of the video camera). The microlenses, because of their respective positions, record different parts of the same scene (as for example a disk and a square spaced laterally), as shown in Figure 2. The microlenses allow a capture of a individual angular information (from different viewpoints of the scene) by directing the rays of light to different parts of the pixel array according to their angle of incidence as shown in Fig. 3. The complete "integral" image thus obtained is of multiple resolution with respect to the resolution of each micro-image. It is therefore important to look for a coding which implements an efficient compression of integral images and / or video streams comprising such integral images. Two coding schemes usually used for the compression of this type of image are cited below.

The first type of coding applies a decomposition of the integral image into several views, followed by a 3D coding technique. The second type of coding applies a 2D coding to the integral image itself. This is a simpler and more direct technique. Nevertheless, it still requires optimizations. In particular, the motion estimation from one image to another can generate, for example, a motion vector pointing to a reference block that is positioned between several micro-images. In this case, the motion vector is not selected, which reduces the efficiency of compression coding. This problem is illustrated in the example of FIG. 4, in which such a reference block is pointed by the arrow F and delimited by clear boundaries. It then appears in this example that the boundaries of consecutive micro-images MLB (dark borders) are included in the reference block, so that the content of this reference block is not continuous on the left-hand side and on the left-hand side. right part of a micro-image (or between a bottom part and a top part by considering an alternative horizontal border of the example of Figure 4). As a result, a current block to be encoded does not correspond to the reference block. It is possible to modify the 2D coding applied to the integral image itself, taking into account particular characteristics of the integral image. For example, it is conceivable to encode an integral image by applying a transform such as a discrete wavelet transform (DWT) followed by a conventional encoding of the transformed coefficients. One can thus predict new prediction modes for encoding the integral images with a 2D codec.

Thus, the problems generally encountered with this type of technique result from the ultra-high resolution that the original integral image (and the associated complexity), as well as the presence of a "grid" of artifacts that form the MLB boundaries. between micro-images, these artifacts making difficult the predictive encoding.

The present invention improves the situation. To this end, it proposes, according to a first aspect, a method of decoding by prediction of a current block of at least one current image, in which a motion vector 15 is obtained pointing to a reference block of a zone of pixels previously decoded, said pixel zone comprising an artifact having a determined position, characterized in that it comprises: determining, as a function of the position of the artifact in said pixel zone, whether the reference block comprises at least a part of the artifact, where appropriate, correcting the reference block by correcting pixels of at least said part of the artifact by extrapolation from other pixels of said pixel area, and - using the block corrected reference for prediction decoding. The aforementioned "current image" may be an integral image as presented above, or alternatively a "multi-view" image, a conventional 2D movie image (eg from an initial recording on a film), etc. such a current image having a reference pixel area containing degradations (claws, spots, or the like). Furthermore, the pixel area "previously decoded" may belong to the current image (for example in "intra" encoding / decoding context) or to another image than the current image (image preceding the current image in a sequence, in "inter" coding for example).

The present invention also relates to a prediction coding method of a current block of at least one current image in which a motion vector pointing to a reference block of a pixel area (which can be coded) is determined. , or initially coded and then decoded), said pixel area comprising an artifact having a determined position. The encoding method comprises the steps of: - determining, as a function of the position of the artifact in said pixel area, whether the reference block comprises at least a part of the artifact, - if necessary, correcting the block reference by correcting pixels at least of said portion of the artifact by extrapolation from other pixels of said pixel area, and using the corrected reference block for coding, by prediction. Of course, in one embodiment, the corrected reference block may first be competed with other reference blocks, during motion estimation, instead of being selected directly for use in encoding. . In this case it is "preselected" 15 at least (to be put in competition). It should therefore be understood that the above terms "use the corrected reference block for coding, by prediction" refer both to an embodiment in which the block is directly selected and an embodiment in which it is first put in competition.

In one embodiment, at the coding and / or decoding, the position of the artifact may be associated with at least one parameter to be transmitted in a code stream, this parameter being obtained at the coding and being used at decoding to identify the position of the artifact in a reference block of a pixel area previously decoded.

Here "parameter obtained" is understood to mean that the parameter can be predefined, in the sense that it is an input parameter of the encoder (and / or decoder) that the encoder reads, or can be determined in the sense that it is detected by a suitable technique (examples of which are described later).

In one embodiment, the current image is an integral image (of the type presented above with reference in particular to FIG. 4), consisting of micro-images separated from each other by grid patterns corresponding to the artifact.

In this embodiment, the corrected pixels are: pixels of a part of the grid pattern that comprises the reference block, and pixels of a part of at least one micro-image, adjacent to a part of at least one other micro-image, said portions being contained in the reference block, the pixels of said at least one other micro-image being used without said correction to constitute the corrected reference block. The micro-image portion whose pixels are to be corrected may be to the right, to the left, above or below the micro-image part that the reference block comprises.

More particularly, the micro-image whose pixels are corrected is a function of: - the direction of the motion vector (vertical, horizontal or diagonal), and - according to a mode of acquisition of the orthoscopic or pseudoscopic integral image, with for a given direction of the motion vector: the pixels of the portion of said at least one micro-image located in a first position with respect to the portion of said at least one other micro-image, to be corrected if the image is pseudoscopic, and - the pixels of the portion of said at least one micro-image located in a second position with respect to the portion of said at least one other micro-image, to be corrected if the image is orthoscopic, the first position being distinct from the second position. In one embodiment, data indicating whether the integral image is orthoscopic or pseudoscopic can be obtained or determined at the encoding and decoding.

In one embodiment, to determine if the reference block comprises at least a part of the grid pattern, the position of the grid pattern in the image is identified by the implementation of at least one of the steps: by searching for correlations in the integral image, by detecting pixels corresponding to grid patterns compared to other pixels of the integral image.

In the sense of the second step above, it can be provided for example that the intensity of the pixels along a line or a column of the integral image shows periodic patterns. The pixels whose intensity is less than a first determined threshold, and / or greater than a second determined threshold, are then associated with the grid pattern.

In addition or alternatively, the position of the gate can be derived from information read from the stream (for example, the size of the micro-lenses of the acquisition device). In one embodiment, pixel correction of the artifact portion in the reference block is accomplished by reference to at least pixels adjacent to the artifact.

In this embodiment, the correction is performed by copying neighboring pixels of the pixels to be corrected. As a variant, the correction can be performed by mask filling corresponding to the pixels 15 to be corrected by an "inpainting" type technique. The present invention also relates to a decoder by prediction of a current block of at least one current image, comprising a processing circuit for obtaining a motion vector pointing to a reference block of a previously decoded pixel zone, said zone 20 pixels having an artifact having a determined position. The processing circuit is arranged to: determine, as a function of the position of the artifact in said pixel area, whether the reference block comprises at least a part of the artifact, if necessary, correct the reference block by correcting pixels at least of said portion of the artifact by extrapolation from other pixels in said pixel area, and using the corrected reference block for prediction decoding. The present invention also relates to a computer program comprising instructions for implementing the above decoding method, when this program is executed by a processor. A schematic algorithm of such a program is illustrated in Figure 5 discussed below. The invention also relates to a non-transitory memory medium storing such a program.

The present invention also provides an encoder by prediction of a current block of at least one current image, comprising a processing circuit for determining a motion vector pointing to a reference block of a previously coded pixel area. said 5 pixel area having an artifact having a determined position. The processing circuit is arranged to: - determine, as a function of the position of the artifact in said pixel area, if the reference block comprises at least a part of the artifact, - if necessary, correct the block of reference by correcting at least pixels of said portion of the artifact by extrapolation from other pixels of said pixel area, and - using the corrected reference block for prediction coding. FIG. 11 discussed further schematically illustrates such encoder and decoder devices, including processing circuits for this purpose (including for example a working memory 15 for temporarily storing the blocks to be corrected, successively, for example on the fly). The present invention also relates to a computer program comprising instructions for implementing the above coding method, when this program is executed by a processor. A schematic algorithm of such a program is illustrated in Figure 6 discussed below. The invention also relates to a non-transitory memory medium storing such a program. As soon as the artefacts have determined or determinable positions in the images (at both encoding and decoding), the decoding in the sense of the invention can be operated in a strictly autonomous manner, independently of the correction of the reference block which has been made to the coding. In practice, the same coding and decoding correction techniques are used. The invention finds an advantageous application to the encoding / decoding of integral images then comprising an artefact having a grid pattern related to the capture of the integral images by an array of microlenses, as illustrated in FIGS. 4, 7a, 7b. , 9 commented below. Moreover, other features and advantages of the invention will become apparent upon examination of the following detailed description, and the accompanying drawings, in which: FIG. 1 illustrates an example of a microwave network device; lenses contiguous to a photodetector array for acquiring integral images such as that shown in FIG. 9; FIG. 2 diagrammatically illustrates the principle of capturing a 3D scene by such a device; FIG. more particularly, the capture of the angle information by such a device; FIG. 4 illustrates an artifact having a MLB grid pattern separating the micro-images from an integral image acquired by a device of the type illustrated in FIG. FIG. 5 illustrates the steps of a decoding method according to one embodiment of the invention; FIG. 6 illustrates the steps of a coding method according to one embodiment of the invention; FIG. 7a illustrates a motion vector (arrow) pointing to a reference block 15 (in dotted lines) in the integral image, here pseudoscopic, the reference block being to be corrected since it includes an artifact corresponding to a part of the FIG. 7b illustrates a motion vector pointing to a reference block in the integral image, here orthoscopic, the reference block being correct since it includes an artefact corresponding to a portion of the grid pattern, FIGS. 8a and 8b show the short uncorrected reference block, respectively brief corrected, in a first exemplary embodiment; FIG. 9 illustrates an integral image obtained by a device of the type shown in FIG. 1; FIG. 10a represents the motion vector (arrow) pointing to the reference block 25 (in dotted lines) in a pseudoscopic integral image, and FIG. 10b illustrates in this case the reference block pixels which are to be corrected, as well as its transformation, - FIG. 11 diagrammatically illustrates encoder and decoder devices, - FIGS. 12a and 12b represent the brief uncorrected reference block and respectively corrected brief *, in a second exemplary embodiment.

FIGS. 5 and 6 are illustrated in which the image processing is decoded and coded, respectively, within the meaning of the invention.

The proposed encoding / decoding solution is applied in a standard "prediction" coding / decoding scheme that uses motion compensation, i.e., which obtains a motion vector mv pointing to a causal zone, that is to say a previous or next already coded / decoded image or a part of the already coded / decoded current image (in "block block copy" coding, based on similarities between blocks image). This causal zone is referred to as the "reference zone" or "reference block" hereafter (bearing the reference "brief" in the drawings). This type of coding / decoding then applies a prediction of a current block with respect to this reference block.

Within the meaning of the invention, the reference block used for the prediction is corrected so as to take into account characteristic artifacts linked, in an exemplary application of the invention, to the capture of the scene by microparticles. lenses. This is particularly in this application artifacts related to the edges of microlenses. The continuity problems related to the passage of a micro-lens to the other are thus solved.

An immediate advantage of the coding technique is that the extrapolated area in the short reference block better predicts the area of a current block b. Therefore, the motion estimation can choose the short block, become relevant for the prediction, and the residual Eb corresponding to the difference between the short-corrected reference block * and the current block b is smaller, and therefore easier. to encode in compression. With reference to FIG. 5, which is examined in parallel with FIG. 7a, during the decoding of the integral image, for a current block b to be decoded (block of solid lines in FIG. 7a), a first step Si is performed. parsing a motion vector mv (arrow in FIG. 7a), pointing to the short reference block in the causal zone (dashed block of dashes on the left in FIG. 7a). Therefore, from the motion vector mv in step S2, the position of the initial reference block is short. It is then practiced a transformation of this brief block, as follows.

In a step S3, the possible artifacts linked here to the MLB boundaries between the micro-images are located in the short block. The MLB boundaries form a grid which can be located by reading its parameters in the data stream. coded (knowledge of parameters such as the positioning and the size of the pattern of the artifact grid, transmitted in the coded stream), or - by detection by algorithmic techniques (for example by automatic identification of the artifacts by evaluation of a image discontinuity, or by identification of the grid by recognizing a general grid shape having right angles between the horizontal and vertical lines and constant steps between the horizontal lines on the one hand and / or between the vertical lines on the other hand, this technique may be preceded by edge detection and the application of a median filter). In the first example above, the parameters can be sent from the encoder to the decoder and be part of the code stream which are then read at decoding in step S3. Possible parameters for defining the grid are, for example: the respective resolutions of the integral image, the lens array, and the micro-images, the width of the grid lines (from one edge to the other) . The pixels whose position is zero (modulo the resolution of a micro-image) are identified as belonging to the grid, and their neighbors contained in an interval corresponding to half the width of a grid line (in pixels ) are also identified as belonging to the grid. In the second example above where these parameters are detected and determined at decoding in step S3, the resolution of a micro-image can be known or calculated by determining the period (in number of pixels) giving a maximum autocorrelation of the image. Then, again, pixels whose position is multiple of the resolution of a micro-image plus or minus a threshold (corresponding to a grid line thickness) are identified as belonging to the grid.

A second means, complementary or alternative, may consist of averaging pixel intensities per micro-image line. Lines whose intensity is below a predetermined threshold are identified as belonging to the grid.

By combining the two techniques of position by correlation and intensity by rows of pixels, the pixels corresponding to these two criteria are then marked as belonging to the grid, and this information can be used in the coding for the extrapolation, and also sent to the decoder for use in decoding.

In a next step S4 of the decoding, the MLB gate pixels possibly present in the short reference block are "extrapolated" to "valid" pixels which are in the vicinity of an MLB boundary. Indeed, it is possible that the short reference block pointed to by the motion vector mv contains one or more MLB boundaries between two or more micro-images. The artifact linked here to the grid pattern entails what can be interpreted as purely illustrative as a picture contrast which: - on the one hand, is erroneous and therefore useless, in terms of information to be coded, and on the other hand, is quite expensive in terms of differences to compress in compression with another image area without necessarily the same artifact.

The MLB boundary pixels are then processed by extrapolation, for example purely by way of copying the neighboring non-black pixels of the black pixels MLB, instead of the black pixels of the MLB grid portion that the block includes. Brief Predictor Examples of processing of the pattern formed by the MLB boundaries have been described above. Reference is now made to FIGS. 8a and 8b to describe the treatment of discontinuities (between micro-images) in the case of a short reference block including at least two distinct micro-image portions. It will be understood that the "correction of artifacts" in a reference block includes, in the broadest sense, the processing of discontinuities between micro-images, as well as the MLB border processing between micro-images (when they are apparent, in particular by black pixels). To avoid discontinuities, the pixels of the next micro-image that "comes after" the MLB boundary in the short reference block are also replaced, as shown in FIGS. 8a and 8b.

In Fig. 8a, the brief initial reference block straddles two consecutive micro-images, and encompasses their boundary. In the example shown in FIG. 8b, the pixels of the right micro-image are kept in the short corrected reference block *. On the other hand, the pixels of the micro-image on the left are replaced, in the same way as those of the 3036016 12 grid. Thus, the short block * is located at the same position as the initial reference block short and then contains the common part from the micro-image on the right, and the estimated part extrapolation (left). In the example given purely by way of illustration in FIG. 8b, the pixel shades at the left edge of the right micro-image are simply reused on extrapolated pixel lines to the left of the short block *. More generally, the extrapolation of the pixels can be carried out by various techniques aiming to reproduce, in continuity, a certain pattern (by using a bilateral filter, or by using automatic correction techniques for artifacts, called " inpainting ", replacing the pixels designated as mask for filling, or still using fractal-based or other techniques). Moreover, according to the embodiment above, the pixels of the right micro-image are extrapolated in the MLB border and in the other micro-image on the left. This realization takes into account a sense of the motion vector mv. Indeed, we take into account the direction of motion vector mv to determine which micro-image to use to "copy" its pixels (by extrapolation) elsewhere in the reference block. With reference again to FIG. 8a, the direction of the motion vector mv (arrow mv) justifies, in the example represented, that the pixels of the left micro-image are extrapolated and the pixels of the right micro-image are kept. On the other hand, in the same example, for a motion vector mv in the opposite direction and in the same short block position, the pixels of the left micro-image are preserved but those of the right micro-image extrapolated the opposite of the situation illustrated in Figure 8b). Thus, the direction of the motion vector mv defines the direction of extrapolation from one micro-image to another in such an embodiment.

More particularly, it is appropriate to distinguish in such an embodiment whether the integral image is orthoscopic (the micro-images being in the same sense as the integral image) or pseudoscopic (the micro-images being inverted with respect to the image). integral, by two axial symmetries, one horizontal and the other vertical). FIG. 9 shows the case of a pseudoscopic integral image from which the micro-images of FIGS. 4, 7a, 8a and 8b and 12a and 12b are drawn. The position of the headband on the hair of the model (top right of Figure 9) is reversed compared to the real integral image.

On the other hand, in FIG. 7b, which is the counterpart of FIG. 7a but here for an orthoscopic image, micro-images appear in the same direction as the integral image. In the general case, the extrapolation is preferably: 5 - in the opposite direction to the motion vector mv for the orthoscopic images, and - in the same direction as that of the motion vector mv for the pseudoscopic images. This choice is explained as follows, with reference to FIG. 10a which illustrates the processing of a pseudoscopic integral image. The vector mv (arrow in dashed lines) points from the current block b (solid lines) to the reference block (in dotted lines) to its left. The position of the object (here a disk) in the micro-images (MIref left, MIref right) pointed by the motion vector is shifted to the left with respect to its position in the current micro-image (MIcurr). The part of the scene that includes the MIref right micro-image 15 contained in the short reference block is therefore also contained in the current block b. The micro-image part MIref left contained in the short reference block is however not contained in the current block b, and must then be replaced by estimated pixel values by extrapolation of the pixel values contained in the micro-image. image MIref right. This results in the short corrected reference block of FIG. 10b (box in solid lines). In the example of FIG. 10b, the extrapolation is still carried out by way of illustration by extension of the pixel shades of the MIref right image, as in the simple example of FIG. 8b. Referring again to FIG. 5, in step S5 the short "extrapolated" reference block is thus obtained in the same position as the short reference block. This brief corrected block * is then used (in place of the initial short block, in the state of the art) to reconstruct at decoded step S6 the decoded block b '. The decoded block is reconstructed from the short corrected reference block and the Eb residue obtained in the received code stream. These steps are repeated until all the decoded blocks b 'are obtained in step S7.

The coded stream that the decoder initially receives may comprise, in addition to the motion vectors mv and the residues Eb, information that the encoder transmits on the particularity of the integral image: orthoscopic or pseudoscopic, and possibly parameters of the 3036016. grid that determines the encoder (position, thickness of the MLB boundaries, not between boundaries, etc.). Nevertheless, in one possible embodiment the information on the feature of the integral: orthoscopic or pseudoscopic image can be determined also at decoding. For example, the motion vectors between each pair of adjacent micro-images of the current integral image can be obtained by block matching. The motion vectors are thus horizontal and vertical.

In the pseudoscopic case, the horizontal motion vectors between a micro-image on the left and a micro-image on the right describe the movement of an object (in a scene of the integral image) between the micro-image on the left and the micro-image on the right and are oriented from left to right. The calculation of these vectors shows that the more a micro-image is on the right, and the more the object of the scene is situated on the right in this micro-image.

Similarly, the vertical motion vectors between a micro-image at the bottom and a micro-image at the top describe the movement of an object between the micro-image at the bottom and the micro-image at the top and are oriented from below. up. The calculation of these vectors shows that the lower a micro-image, and the lower the object is in this micro-image.

In the orthoscopic case, the horizontal motion vectors between a micro-image on the left and a micro-image on the right describe the movement of an object between the micro-image on the left and the micro-image on the right and are oriented from right to left. The calculation of these vectors shows that the more a micro-image is on the right, and the more the object of the scene is situated on the left in this micro-image.

The vertical motion vectors between a micro-image at the bottom and a micro-image at the top describe the movement of an object between the micro-image at the bottom and the micro-image at the top and are oriented from top to bottom. The calculation of these vectors shows that the lower a micro-image, and the higher the object is located in this micro-image.

Referring now to FIG. 6, to the coding, for each possible candidate vector when searching for motion in step CS1, a motion vector mv is obtained pointing to a short reference block at step CS2.

In general, the processes presented above for the decoding with reference to FIG. 5, which make it possible to determine the position of the MLB boundaries (parameters of the grid such as the position, the thickness of the MLB boundaries, the not between the borders, etc.), as well as possibly the particularity of the integral: orthoscopic or pseudoscopic image, without receiving this data at the decoder, can be done similarly to the coding. Thus, with reference to FIG. 6, the following step CS3 to the coding comprises the location of the MLB boundaries forming the gate, and in particular in the short reference block to be processed.

The next step CS4 includes the extrapolation of the micro-image pointed by the motion vector mv, to create the short extrapolated predictor block at step CS5, at the same position as the brief initial block. The following step CS6 conventionally comprises calculating the Eb residue by subtracting the short corrected predictor block from the current block to be coded b. Optionally, there may be an additional step CS7 comprising selecting the best motion vector mv by estimating a cost function FC. Finally, the FLC coded stream is obtained at step CS8 comprising at least the motion vectors and the residual blocks. The coded stream may also initially comprise, for example, information data in which the image is orthoscopic or pseudoscopic, and possibly information on the grid parameters (position, thickness, etc.). Coding as well as decoding, a SOK step of selecting the implementation or not of the treatment of the invention may be provided. Indeed, in order to save computing resources, the activation of the treatment of the invention can be carried out automatically according to one or more of the following activation criteria: the size of the zone to be extrapolate must remain small (otherwise, the processing is not done), - the size of the motion vector mv must be less than a threshold, otherwise there is no chance of finding missing pixel data in the distant micro-image that points the motion vector.

Moreover, the processing can be applied to all the blocks if the above criterion or criteria are met, but need not be implemented particularly in cases where the short reference block does not need to be used. contains no MLB boundary (information obtained by viewing grid position data). On the other hand, if the current block contains a horizontal or vertical MLB boundary, then a motion vector mv itself horizontal or vertical is chosen. Nevertheless, the motion vector can be diagonal, which implies an extrapolation that is not restricted to a single dimension. FIG. 11 shows an encoder device as well as a decoder device within the meaning of the invention. The coder COD comprises a processing circuit including for example: an input ENT1 for receiving the blocks b of images to be encoded, processing means MT1 of the blocks b, comprising for example a processor PROC1 and a working memory MEM1, to apply the method of the invention for the coding of the blocks with an extrapolation of pixels in the reference blocks 15 brief *, and - an output SOR1 to deliver the FLC coded stream resulting from the coding of the blocks. The DECOD decoder comprises a processing circuit including, for example: an input ENT2 for receiving the FLC coded stream, MT2 processing means of the FLC stream, comprising for example a processor PROC2 and a working memory MEM2, for applying the method of the invention for the decoding of the FLC flux and a correction of the short reference blocks by extrapolation of their pixels, and - an output SOR2 to deliver the blocks b 'of decoded and available images for image restitution.

Of course, the present invention is not limited to the embodiments described above by way of example; it extends to other variants. Typically, an application of the invention to integral images obtained from a microlens array has been described above. However, it applies to any type of image in which at least one artifact affecting at least one reference block is to be corrected to ensure effective compression. It should also be noted that the redundancy of the micro-images (specific to an integral image) is not used in the treatment examples above.

Thus, an integral image comprising a multiplicity of micro-images is not an essential characteristic for the implementation of the invention, the essential point being to have a determination of the position of the artifact to be deleted in the reference block to correct.

Furthermore, in FIG. 4 in particular, there appears a pattern of artifacts that the MLB boundaries form between micro-images in the form of a "grid" of horizontal and vertical lines. Nevertheless, with acquisition devices having lenses of different shapes (hexagonal, spherical, square, or other) or of different sizes ("RaytrixTM" device), or arranged in staggered rows rather than aligned, the pattern 10 d artifacts can have a different appearance (honeycomb for example, or other), the main processing steps according to the invention are not modified so far (search for this particular form of pattern, correction of the pixels in the identified regions).

Claims (15)

REVENDICATIONS1. A method of decoding by prediction of a current block of at least one current image, in which a motion vector (mv) is obtained pointing to a reference block (short) of a previously decoded pixel area, said area of pixels comprising an artifact having a determined position, characterized in that it comprises: - determining (S3), as a function of the position of the artifact in said pixel area, if the reference block comprises at least a part of the artifact, - where appropriate, correcting (S4) the reference block by correcting pixels of at least said part of the artifact by extrapolation from other pixels of said pixel area, and - using the reference block corrected (short) for decoding (S6) by prediction. 2. Method for coding by prediction of a current block of at least one current image in which a motion vector (mv) pointing to a reference block (short) of a pixel area, said pixel area is determined comprising an artifact having a determined position, characterized in that it comprises: - determining (CS3), as a function of the position of the artifact in said pixel area, if the reference block comprises at least a part of the artifact, - where appropriate, correcting (CS4) the reference block by correcting pixels of at least said part of the artifact (CS4) by extrapolation from other pixels of said pixel area, and - using the block corrected reference (brief) for coding (CS6), by prediction. 3. Method according to one of claims 1 and 2, characterized in that the position of the artifact is associated with at least one parameter to be transmitted in a coded stream, said at least parameter being obtained coding and being used decoding to identify the position of the artifact in a reference block of a previously decoded pixel area. 4. Method according to any one of claims 1 to 3, characterized in that the current image is an integral image, consisting of micro-images separated from each other by grid patterns (MLB) corresponding to the artifact. 3036016 19 5. Method according to claim 4, characterized in that the corrected pixels are: pixels of a part of the grid pattern that comprises the reference block, and pixels of a part of at least one micro-pixel. image, adjacent to a portion of at least one other micro-image, said portions being contained in the reference block, the pixels of said at least one other micro-image being used without said correction to constitute the corrected reference block (short). 6. Method according to claim 5, characterized in that the micro-image whose pixels are corrected is a function of: - the direction of the motion vector, and - according to a mode of acquisition of the orthoscopic or pseudoscopic integral image with, for a given direction of the motion vector: - the pixels of the part of said at least one micro-image located in a first position with respect to the part of said at least one other micro-image, to be corrected if image is pseudoscopic, and - the pixels of the part of said at least one micro-image located in a second position with respect to the part of said at least one other micro-image, to be corrected if the image is orthoscopic the first position being distinct from the second position. 20 7. Method according to claim 6, characterized in that a data item indicating whether the integral image is orthoscopic or pseudoscopic is determined at the coding and decoding. 8. Method according to one of claims 4 to 7, characterized in that, in order to determine if the reference block comprises at least a part of the grid pattern, the position of the grid pattern in the image is identified by the implementing at least one of the steps: - by searching for correlations in the integral image, - by detecting pixels corresponding to grid patterns compared to other pixels of the integral image. 30 9. Method according to one of the preceding claims, characterized in that the pixel correction of the artifact portion in the reference block is performed with reference to at least pixels adjacent to the artifact. 3036016 20 10. The method of claim 9, characterized in that the correction is performed by copying neighboring pixels of the pixels to be corrected. 11. Method according to claim 9, characterized in that the correction is performed by mask filling corresponding to the pixels to be corrected by an inpainting technique. 12. Decoder by prediction of a current block of at least one current image, comprising a processing circuit (PROC2, MEM2) to obtain a motion vector (mv) pointing to a reference block (brief) of a previously decoded pixel area, said pixel area having an artifact having a determined position, characterized in that the processing circuit is arranged to: - determine, as a function of the position of the artifact in said pixel area, whether the reference block comprises at least a part of the artifact, 15 - where appropriate, correcting the reference block by correcting pixels of at least said part of the artifact by extrapolation from other pixels of said pixel area , and - use the corrected reference block (brief) for prediction decoding. 13. Computer program characterized in that it comprises instructions for implementing the decoding method according to one of claims 1, 3 to 11, when the program is executed by a processor. 14. Coder by prediction of a current block of at least one current image, comprising a processing circuit (PROC 1, MEM1) for determining a motion vector (mv) pointing to a reference block (brief) of a previously coded pixel area, said pixel area comprising an artifact having a determined position, characterized in that the processing circuit is arranged to: determine, as a function of the position of the artifact in said pixel area, whether the reference block comprises at least a part of the artifact, if necessary, correcting the reference block by correcting pixels at least of said part of the artifact by extrapolation from other pixels of said pixel area, and use the corrected reference block (brief) for prediction coding. 3036016 21 15. Computer program characterized in that it comprises instructions for implementing the coding method according to one of claims 2 to 11, when the program is executed by a processor.