FR2966681A1 - Image slice coding method, involves determining lighting compensation parameter so as to minimize calculated distance between cumulated functions, and coding image slice from reference image - Google Patents

Abstract

The method involves determining a covering zone (10) between a current image slice and a reference image, and calculating a histogram (12) of a part of the current image slice in the covering zone and another histogram of a part of reference image in the covering zone. Cumulated functions of the histograms are calculated (14). A lighting compensation parameter is determined (16) so as to minimize calculated distance between the cumulated functions of the histograms balanced by the lighting compensation parameter. The image slice is coded (18) from the reference image.

Description

METHOD OF ENCODING IMAGE SLICE

FIELD OF THE INVENTION The invention relates to the general field of image coding.

The invention relates more particularly to a method of encoding an image slice from at least one reference image.

2. State of the art Most block coding / decoding methods use inter-image prediction (inter-image prediction or inter mode) or image prediction (intra-image prediction or intra mode). Such a prediction makes it possible to improve the compression of a sequence of images. The inter prediction consists in generating a prediction block from a block of a reference image (previously coded) for a current block to be coded and coding the difference between the current block and the prediction block, also called residual block. The more the prediction block is correlated to the current block, the smaller the number of bits needed to code the current block and thus the effective compression. However, the prediction loses its effectiveness when there is a variation in brightness between the images of the sequence or within an image. Such a change in brightness is for example due to an illumination modification, fade effects, flashes, etc. There are known image sequence coding / decoding methods that take into account an overall variation in brightness. Thus, in the context of the H.264 standard described in ISO / IEC 14496-10, it is known to use a weighted prediction method to improve the compression in the case of a variation in brightness. The illumination compensation parameters associated with the reference images (known as "weighted prediction parameters") are explicitly transmitted by image slice (also known as "slice"). The illumination correction applies identically for all blocks in the slice. Such parameters are for example determined from the DC component (DC coefficient) of a current block and the DC component of a reference block associated with the current block. Such a method makes it possible to determine that a single parameter is additive or multiplicative. According to one variant, a multiplicative parameter and an additive parameter are determined by minimizing the mean squared error calculated between a current image and a reference image associated with the current image. Such a solution is described in the paper by Kato et al entitled "Weighted factor determination algorithm for H.264 / AVC weighted prediction" and published in 2004 in the Proceedings of the Workshop on Multimedia Signal Processing. Such a method is, however, very sensitive to motion because the error is calculated pixel by pixel between the pixels of the current image and the co-located pixels in the reference image. According to another approach described in Shen et al entitled "Adaptive Weighted Prediction in Video Coding" published in 2004 in the proceedings of the ICME conference, a motion vector and parameters are determined by minimizing the mean squared error calculated between a current image and a reference image associated with the current image compensated in motion. Such an approach has a very high computational cost. SUMMARY OF THE INVENTION The object of the invention is to overcome at least one of the drawbacks of the prior art. For this purpose, the invention relates to a method of coding an image slice, called the current slice, from at least one reference image comprising the following steps: determining at least one illumination compensation parameter ; and - coding the current slice from the reference image taking into account the determined illumination compensation parameter. Advantageously, the step of determining the illumination compensation parameter comprises the following steps: determining an overlap zone between the current slice and the reference image; calculating a histogram of the portion of the current slice in the overlapping zone, said first histogram, and a histogram of the portion of the reference image in the overlapping zone, said second histogram; calculating a cumulative function of the first histogram, called the first cumulative function, and a cumulative function of the second histogram, called the second cumulative function; determining the illumination compensation parameter so as to minimize a calculated distance between the first cumulative function and the second cumulative function weighted by the illumination compensation parameter. The method according to the invention advantageously allows a precise and robust determination of the illumination compensation parameters because it is not based on a difference between co-located pixels but on histograms of images. It is therefore not sensitive to local or global movement. In addition, it is fast because precisely it is based on the calculation of image histograms. No local movement compensation is necessary.

According to a particular embodiment, a multiplicative illumination compensation parameter and an additive illumination compensation parameter being determined, the current slice is coded from the reference image taking into account the illumination compensation parameters. when the calculated distance for the determined illumination compensation parameters is less than a first predetermined threshold and when the ratio between the calculated distance for the determined illumination compensation parameters and the calculated distance for a multiplicative parameter equal to 1 and a additive parameter equal to zero is less than a second predetermined threshold and in which the current slice is encoded from the reference image without taking into account illumination compensation parameters otherwise. According to a particular aspect of the invention, the overlap zone is determined by registering the current slice with the reference image. According to an alternative embodiment, the overlap zone is determined by global motion compensation of the current slice relative to the reference image from a global motion vector. Advantageously, the global motion vector is the average motion vector computed from motion vectors associated with blocks of the slice of the reference image co-located with the current slice, the motion vectors being rescaled to take into account the temporal distance between images. Advantageously, the global motion vector is the median motion vector calculated from motion vectors associated with blocks of the slice of the reference image co-located with the current slice, the motion vectors being rescaled to take into account the temporal distance between images. According to another variant embodiment, the overlap area is determined by global motion compensation of the reference image from parameters representative of a camera movement between the reference image and the current image.

4. Lists of Figures The invention will be better understood and illustrated by means of examples of advantageous embodiments and implementations, in no way limiting, with reference to the appended figures in which: FIG. 1 illustrates a coding method according to the invention; FIG. 2 represents a current image and a reference image as well as a zone of overlap of the two images; FIG. 3 represents a cumulative function of the histogram of a current image and a cumulative function of the histogram of a reference image; and FIG. 4 represents a coding device according to the invention.

5. Detailed Description of the Invention In the following, the term "coding entity" designates a basic structure of an image to be coded or reconstructed. It comprises a subset of contiguous pixels of the image, i.e. not disjoint. The set of coding entities of an image may or may not have common pixels. Generally, the encoding entity is a block of pixels. However, the term "coding entity" is generic and may designate a square, a rectangle, a circle, a polygon, a region of any shape. A block of pixels is a particular encoding entity. The term "prediction entity" refers to the basic structure used for the prediction of a coding entity. It comprises a subset of contiguous, i.e. non-disjoint pixels. Generally, the encoding entity is a block of pixels.

In the literature, the prediction entity is generally a block. However, the term "prediction entity" is generic and can mean a square, a rectangle, a circle, a polygon, a region of any shape. A prediction entity is used to predict an encoding entity, p.e.x, for calculating a residual error. An image slice is formed of a plurality of encoding entities. With reference to FIG. 1, the invention relates to a method of encoding a current Scur slice of a current image Icur from reference images Iref ;, where i is the index of the reference image. Current and reference images belong to a sequence of images. In the same image slice, some blocks can use the image Iref; as a reference image and other blocks can use the image Iref as a reference image with j i. In the remainder of the description, the term block is used for the sake of simplification. However, this term can be replaced by the generic term coding entity. In a step 10, a recovery area REC is determined between the current Scur slice to be encoded and a reference image Iref; This overlap area REC corresponds to the parts of the scene common between the current slice and the reference image. Such a recovery zone REC is shown hatched in FIG. 2. This overlapping zone is, for example, determined from an image registration algorithm. An example of such an algorithm is described in P.Saravanan et al., Entitled "Techniques for Video Mosaicing", published in World Academy of Science, Engineering and Technology in 2005. Alternatively, the area of overlap REC is determined by global motion compensation of the current slice relative to the reference image Iref; This global motion compensation is performed from a GMV global motion vector. This global motion vector is, for example, equal to the vector obtained by averaging the motion vectors mv associated with blocks B of the slice of the reference image Iref; collocated to the current slice, said motion vectors being scaled to account for the temporal distance between images. The collocated slice is the slice that occupies the same place in the reference image Iref; that Scur in the current picture Icur. With reference to FIG. 3, the scaled motion vector mv is therefore equal to (Tcur-Tref;) / (Tref; -T,) * mv. These motion vectors are the motion vectors used during the coding of the reference image Iref; In the particular case where all the blocks of the reference image Iref; are encoded in intra mode, with or without spatial prediction, then the GMV motion vector is a vector defined by default, eg the last determined GMV vector or the null vector. Similarly, in the case where the motion vector GMV leads to the absence of a recovery zone, it is replaced by a vector defined by default, eg the null vector. According to one variant, this global motion vector GMV is the median vector of the motion vectors associated with blocks of the slice of the reference image Iref; collocated to the current slice, said motion vectors being scaled to account for the temporal distance between images. According to another variant, the global motion compensation is performed from parameters representative of a camera movement between the reference image and the current image, the camera having served to capture the sequence of images. In a step 12, a histogram Hist1, said first histogram, of the portion of the current slice Scur in the overlap area and histogram Hist2, said second histogram, of the portion of the reference image Iref; in the overlap area are calculated in the overlap area.

During a step 14, a cumulative function L1, called the first cumulative function, of the first histogram and a cumulative function, called the second cumulative function L2, of the second histogram are calculated. Such functions are shown in FIG. 4. In a step 16, one or more illumination compensation parameter (s) is (are) determined so as to minimize a distance dist (w; calculated between the first cumulative function and the second cumulative function weighted by the illumination compensation parameter (s). The distance is for example equal to n = 100% dist (wi, oi) = (L1 (n) - wL x L2 (n) - oi) zn = 0% where (w;, o;) are the compensation parameters of illumination, L1 (n) is the abscissa corresponding to the value "n" of the cumulative function L1 and L2 (n) is the abscissa corresponding to the value "n" of the cumulative function L2. According to a variant, the distance is equal to dist (w, o) = dist (wi, oL) = En = yr1 L1 (n) - wi x L2 (n) - oil In a particular embodiment, a conventional minimization method square least is used. In this case, the parameters (w;, o;) are those which satisfy the following system of equations: ## EQU1 ## Steps 10 to 16 are repeated for all the reference images Iref; usable to code the blocks of the current Scur slice. In a step 18, the blocks of the current slice are coded from one of the reference images taking into account the illumination compensation parameter (s) determined in step 16. Thus, for a current block of the current slice, the residual error between this current block and a motion compensated reference block weighted by the illumination compensation parameters determined in step 16 is calculated. This residual error is therefore equal to Icur (x, y) -w; * Iref; (x + Ax, y + Ay) -o; where (Ax, Ay) are the coordinates of the motion vector associated with the block to be encoded, where (x, y) are the coordinates of the pixels of the current block to be encoded and where (w;, o;) are the compensation parameters of illumination determined from the reference image Iref; Generally, the residual errors calculated for a block are transformed into coefficients for example with a DCT (acronym for Discrete Cosine Transform) then quantized before their entropy coding. According to an alternative embodiment of step 18, the illumination compensation parameters are not systematically applied. Indeed, the use of such parameters is not always appropriate especially in the case of scene change or in the case of local variation of illumination. During a step 17, prior to the coding step 18, it is determined whether it is appropriate to use the illumination compensation parameters determined and that for each reference image Iref; used to encode blocks of the current slice. For this purpose, the illumination compensation parameters associated with a reference image Iref; are used to encode the blocks of the current slice using the Iref image; as a reference image if the following inequalities are satisfied: dist (w;, o;) <TH1 (1) and dist (w;, o;) / error (1,0) <TH2 (2), where TH1 and TH2 are two threshold values. In the opposite case, the weighting parameters (w;, o;) are not used for the coding of the blocks of the current slice that use the image Iref; as a reference image, i.e. the residual error is equal to Icur (x, y) - Iref (x + Ax, y + Ay) where (Ax, Axy) are the coordinates of the motion vector associated with a block to be encoded.

In the case where the parameters (w;, o;) are not used to code the blocks of the current slice using Iref; as a reference image, a flag is encoded in the stream, more precisely in the header of the Scur slice, indicating that for the reference image Iref; no illumination compensation parameter is transmitted, and therefore consequently used. In the case where the parameters (w;, o;) are used to code the blocks of the current slice using Iref; as a reference image, a flag is encoded in the stream, more precisely in the header of the Scur slice, indicating that for the reference image Iref; illumination compensation parameters are transmitted, said parameters being encoded in the stream as a result of this flag. According to another variant embodiment, for the blocks of the current slice which use as reference image an image Iref; for which illumination compensation parameters are transmitted, a flag is coded in the header of these blocks indicating whether the parameters (w;, o;) are actually used to code and thus reconstruct this block. By way of illustrative example, a Scur image slice comprises 4 blocks B1, B2, B3 and B4 which are coded with reference to 2 reference images Iref, and Iref2. Only the illumination compensation parameters (w ,, o,) are transmitted in the header of the slice. The illumination compensation parameters (w2, 02) determined in step 16 for the reference image Iref2 are not transmitted because they do not check the inequalities (1) and (2). Assume that blocks B1 and B3 are encoded with reference to image Iref ,, then an additional flag is coded in the header of each block to indicate whether the block in question actually uses the parameters (w ,, o,) in the header of the Scur slice. For example, a block Bk uses the illumination compensation parameters (w;, o;) if the following condition is satisfied:

Pelcur (x, y) - wL X Pelre fi (x + Ox, y + Dy) - oL (x, y) EBk <I Pelcur (x, y) - PelYefi (x + Ox, y + oy) (x, y) EBk where Pelcur (x, y) is the value of the pixel of coordinates (x, y) in the block Bk and Pelrefi (x + Ax, y + Ay) is the value of the corresponding pixel in the reference image I refi. Assume that blocks B2 and B4 are encoded with reference to image Iref2, then they are encoded without regard to any illumination compensation parameter.

The invention also relates to a coding device 12 described with reference to FIG. 5. The coding device 12 receives images 1 belonging to a sequence of images as input. Each image is divided into blocks of pixels each of which is associated with at least one image data, e.g. luminance and / or chrominance. The coding device 12 implements, in particular, coding with temporal prediction. Only the modules of the coding device 12 relating to the coding by temporal prediction or inter coding are represented in FIG. 12. Other modules, not shown and known to those skilled in the video coder art, implement the intra coding with or without spatial prediction. The coding device 12 comprises in particular a calculation module ADD1 able to subtract pixel by pixel from a current block Bc a prediction block Bp to generate a block of residual image data or residual block denoted res. 11 further comprises a TQ module adapted to transform and quantify the residual block res in quantized data. The transform T is for example a discrete cosine transform (or "DCT"), which stands for "discrete cosine transform". The coding device 12 further comprises an entropic coding module COD able to encode the quantized data into a stream F of coded data. It further comprises an ITQ module performing the inverse operation of the TQ module. The ITQ module performs an inverse quantization Q- 'followed by an inverse transformation T-'. The ITQ module is connected to a calculation module ADD2 capable of adding pixel by pixel the data block from the ITQ module and the prediction block Bp to generate a block of reconstructed image data which are stored in a memory MEM. The coding device 12 furthermore comprises a motion estimation module ME able to estimate at least one motion vector between the block Bc and a block of a reference image Irefi stored in the memory MEM, this image having been previously coded and rebuilt. According to a variant, the motion estimation can be made between the current block Bc and the original reference image in which case the memory MEM is not connected to the motion estimation module ME. According to a method well known to those skilled in the art, the motion estimation module looks in the reference image Iref; a motion vector so as to minimize an error calculated between the current block Bc and a block in the reference image Iref; identified using said motion vector. The motion data is transmitted by the motion estimation module ME to a decision module DECISION able to select an encoding mode for the block Bc in a predefined set of coding mode. The coding mode chosen is for example the one that minimizes a criterion of the type of bitrate-distortion. However, the invention is not limited to this selection method and the selected mode can be selected according to another criterion, for example a criterion of the prior type. The encoding mode selected by the decision module DECISION as well as the motion data, eg the motion vector or vectors in the case of the temporal prediction mode or the inter mode are transmitted to a prediction module PRED. The motion vector or vectors and the selected coding mode are furthermore transmitted to the entropic coding module COD in order to be coded in the stream F. If an inter prediction mode is retained by the decision module DECISION, the prediction module PRED then determines in the reference image Iref; previously reconstructed and stored in the memory MEM, the prediction block Bp from the motion vector determined by the motion estimation module ME and the coding mode determined by the DECISION decision module. If a prediction mode INTRA is retained by the decision module DECISION, the prediction module PRED determines in the current image, among the previously coded blocks and stored in the memory MEM, the prediction block Bp.

The prediction module PRED is able to determine the prediction block Bp taking into account a brightness variation model defined by illumination compensation parameters also called parameters in the particular context of the prediction weighting image coding. representative of a variation of brightness between the images of the sequence or within an image.

For this purpose, the coding device 12 comprises a module for determining illumination compensation parameters IC. The determination module IC is able to implement steps 10 to 16 of the coding method described with reference to FIG.

Of course, the invention is not limited to the embodiments mentioned above. In particular, those skilled in the art can make any variant in the exposed embodiments and combine them to benefit from their various advantages.

In particular, the invention is in no way limited to blocks of pixels, but applies to coding entities as previously defined. Moreover, the invention defined for two parameters the one multiplicative (w;) the other additive (o;), is not limited to two parameters and can be applied to an illumination compensation model involving only one of the two parameters or a more complex model involving other illumination compensation parameters.

Claims (7)

REVENDICATIONS1. A method of coding an image slice, called the current slice, from at least one reference image comprising the following steps: - determining (16) at least one illumination compensation parameter; - coding (18) said current slice from said reference image taking into account said determined illumination compensation parameter; said coding method being characterized in that the step of determining said illumination compensation parameter comprises the following steps: - determining (10) an overlap area between said current slice and said reference picture; calculating a (12) histogram of the portion of the current slice in the overlapping zone, said first histogram, and a histogram of the portion of the reference image in said overlapping zone, said second histogram; calculating (14) a cumulative function of the first histogram, called the first cumulative function, and a cumulative function of the second histogram, called the second cumulative function; determining (16) said illumination compensation parameter so as to minimize a calculated distance between the first cumulative function and the second cumulative function weighted by said illumination compensation parameter. An encoding method according to claim 1, wherein, a multiplicative illumination compensation parameter and an additive illumination compensation parameter being determined (16), said current slice is encoded from said reference image by taking counting said illumination compensation parameters when the calculated distance for said determined illumination compensation parameters is less than a first predetermined threshold and when the ratio between said calculated distance for said determined illumination compensation parameters and said calculated distance for said a multiplicative parameter equal to 1 and an additive parameter equal to zero is less than a second predetermined threshold and wherein said current slice is coded from said reference image without taking into account said illumination compensation parameters otherwise. 3. Encoding method according to one of the preceding claims, wherein said overlap area is determined by resetting said current slice with said reference image. 4. coding method according to one of claims 1 to 2, wherein said overlap area is determined by global motion compensation of said current slice relative to said reference image from a global motion vector. The coding method according to claim 4, wherein said global motion vector is the average motion vector computed from motion vectors associated with blocks of the slice of the reference image co-located with the current slice, said motion vectors being scaled back to account for the temporal distance between images. The coding method according to claim 4, wherein said global motion vector is the median motion vector computed from motion vectors associated with blocks of the slice of the reference image co-located with the current slice, said motion vectors being scaled back to account for the temporal distance between images. The encoding method according to one of claims 1 to 2, wherein said overlapping area is determined by global motion compensation of said reference image from parameters representative of a camera movement between the reference image. and the current image