Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
  • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
  • H04N19/51—Motion estimation or motion compensation
  • H04N19/513—Processing of motion vectors
  • H04N19/517—Processing of motion vectors by encoding
  • H04N19/52—Processing of motion vectors by encoding by predictive encoding
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
  • H04N19/593—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial prediction techniques
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
  • H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
  • H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
  • H04N19/51—Motion estimation or motion compensation
  • H04N19/57—Motion estimation characterised by a search window with variable size or shape

Definitions

• the present invention relates generally to the field of image processing, and more specifically to the coding and decoding by competition of digital images and digital image sequences.
• coding and decoding methods exist for the transmission of images.
• major types of coding such as the so-called “intra” coding where an image is coded autonomously, that is to say without reference to other images, or the so-called “inter” coding which consists of to encode a current image with respect to past images so as to express and transmit only the difference between these images.
• Coding methods of the aforementioned type generally comprise a predictive coding step in which portions of images, called blocks or macroblocks, of a current image are predicted with respect to other reference blocks or macroblocks, that is, ie previously coded and decoded.
• the predictive coding of a macroblock is to split the macroblocks into a plurality of partitions generally having the shape of smaller blocks.
• the current macroblock to be encoded can be partitioned according to the 16x16, 8x16, 16x8 and 8x8 modes. If the 8x8 mode is selected, each 8x8 block is partitioned again according to the 8x8, 4x8, 8x4 and 4x4 modes. Each current block is compared to one or more blocks respectively of one or several reference images. A motion vector is then obtained which describes the movement between the current block and the reference block having the same position as the current macroblock in the previous image. A predictor of this motion vector is then calculated to encode the residual between the aforementioned motion vector and the calculated predictor motion vector.
• Such a motion vector prediction is not suitable for all types of partitioning.
• FIG. 1A represents, for example, the case of a temporal prediction of an MVpI motion vector of a partition P1 of a current macroblock to be encoded, denoted MBCN, which belongs to an image N to be coded according to the aforementioned standard.
• MBCN macroblock is of conventional square shape.
• the partition P1 which is smaller than that of the current macroblock MBC N , also has a square shape.
• the partition P1 is also surrounded by other macroblocks BR1, BR2, BR3, BR4 which are located in the close vicinity of the latter and which, in the example shown, have the same shape and size as those of the macroblock MBC N.
• the motion vector MVpI points to a zone p'1 of a reference image, denoted N-1, which is for example the immediately preceding image.
• the reference image area p'1 has the same position as the partition p1 of the current macroblock MBCN in the previous picture N-1 and is close to a plurality of reference partitions r'1, r'2, r '3 and r'4.
• MVpI supra is only predicted spatially. More precisely, a reference motion vector is computed which is equal to a median of the motion vectors MV1, MV3, MV4 associated respectively with reference macroblocks BR1, BR3 and BR4.
• a current macroblock to be encoded MBC N can be divided into several partitions P1 to Pp of linear form, L-shaped, or of quite arbitrary form.
• the H264 / AVC standard does not provide for a motion vector prediction that is adapted to the different types of partitioning of FIG. 1B and to the particular case where the reference image area pointed by this vector has a geometrical shape or a different size. respectively the geometric shape or the size of at least one neighboring reference partition.
• FIG. 1C illustrates the case of a prediction of several motion vectors MVpI, MVp2, MVp3 associated respectively with three partitions P1, P2, P3 of a current macroblock to be coded MBCN belonging to an image N to code in accordance with the above standard.
• the partitions P1, P2 and P3 of the macroblock MBCN are of any geometric shape.
• the motion vector of the first partition P1 of the current macroblock MBCN points to a reference image area p'1 of a reference macroblock MBCN- I of a reference image N -1 having the same position as the current macroblock MBC N in the previous image N-1.
• a reference partition r'1 close to the reference image area p'1 has a geometrical shape very different from that of the reference image area p'1 and the standard H264 / AVC does not propose a calculation method specific to such a case.
• Such a spatial prediction of the motion vector may be unclear given the fact that in the N-1 image there is a difference in shape and size between the reference picture area p'1 and the reference partitions r'1, r'2, r'3, r'4.
• FIG. 1D represents, for example, the case of a spatial prediction of an MVpI motion vector of a partition P1 of a current macroblock to be encoded, denoted MBC N , which belongs to an image N to be coded according to the above-mentioned standard .
• the conventional prediction method used in the H264 / AVC standard consists in calculating the MVpI motion vector of the partition P1 as the median of the reference motion vectors respectively associated with the reference partitions neighboring the partition P1, namely the partitions BR1, BR3 and BR4 in FIG.
• the conventional prediction method will therefore not be adapted to the particular partitioning illustrated in FIG. 1 D. Indeed, such a method does not impose rules for selecting the most important reference partitions. adapted in terms of geometric shape or size, and therefore corresponding reference motion vectors for the prediction of MVpI.
• the motion vector of a macroblock of a current image is predicted with respect to a reference motion vector which is chosen to be the vector pointing to the pixel located in top and leftmost of the macroblock having the same position as the current macroblock in a previous image.
• the vector MVpI of FIG. 1A is obtained from a reference motion vector which is associated with the reference partition r'2, the leftmost pixel of the reference partition p'1 being located in the partition reference r'2;
• the vector MVpI of FIG. 1C is obtained from a reference motion vector which is associated with the reference partition r'1, the leftmost pixel of the reference partition p'1 being located in the partition reference r'1.
• the prediction obtained with this coding method is also not adapted to the different types of partitioning for the same reasons as those mentioned above in connection with the prediction method used in the H264 / AVC standard.
• One of the aims of the invention is to overcome disadvantages of the state of the art mentioned above.
• the present invention relates to a method for spatially prediction of a motion vector of a partition of a current image, from a plurality of n reference motion vectors respectively associated with n partitions. of the current image that have been previously coded and decoded.
• the motion vector of the current image partition is determined from a function at least one reference motion vector which belongs to a set of k reference motion vectors respectively associated with the k neighboring reference partitions.
• the prediction according to the invention is adaptable to any type of calculation method of the predicted motion vector of the current partition, such as in particular that conforming to the H264 / AVC standard and that described in the aforementioned IEEE publication.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select very specific reference motion vectors from a selection criterion which is here based on the proximity between the current partition on the one hand and the partition (s) on the other ( s) the nearest one (s).
• determining the motion vector of the current partition comprises the steps of:
• the determination of the motion vector of the current partition comprises the steps of:
• Such arrangements thus make it possible to select even more finely and more precisely one or more reference motion vectors from a selection criterion which is here based not only on the proximity between the current partition and the partition (s) ( s) nearest reference (s), but also on a comparison of the distance between on the one hand the center of the current partition and, on the other hand, the center of each of the neighboring reference partitions.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select in a targeted manner one or more very specific reference motion vectors from a selection criterion which is here based on the existence of a discontinuity in the previously coded macroblocks, then decoded.
• the accuracy of the prediction is thus improved because it is obtained from reference motion vectors that are supposed to be more accurate in an area of the image that contains discontinuities than in a zone of the homogeneous image.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select very specific reference motion vectors from a selection criterion which is here based on the proximity between the current partition on the one hand and the partition (s) on the other ( s) the nearest one (s).
• Such a selection is further refined by determining the longest edge of the current partition which delimits the latter of at least one neighboring reference partition.
• the selection step is followed by a step of calculating the average of the kb of the reference movement vectors respectively corresponding to the kb of selected reference partitions.
• the selection step consists in choosing, in the subset of k ⁇ reference partitions having said longest edge in common, the reference partition whose edge portion in common is the longest, then the reference movement vector corresponding to the selected reference partition.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select a very precise reference motion vector from a certain characteristic which is here based on the identification of a particular pixellic component, such as for example the average luminance of a pattern, of a color, an outline, etc. in a reference partition adjacent to the current partition.
• the present invention relates to a method of coding an image or a sequence of images generating a data stream comprising data representative of at least one image partition, such a method comprising a step spatial prediction of a motion vector of the image partition.
• Such a coding method is characterized in that the spatial prediction is performed in accordance with the aforesaid spatial prediction method.
• the present invention relates to a method of decoding a data flow representative of an image or a sequence of images, said flow comprising data representative of least one image partition, such a method comprising a step of spatially prediction of a motion vector of the image partition.
• Such a method is characterized in that the spatial prediction is performed in accordance with the aforesaid spatial prediction method.
• the present invention relates to a device for spatial prediction of a motion vector of a partition of a current image, from a plurality of n reference motion vectors respectively associated with n reference partitions. of the current image that have been previously coded and decoded.
• the spatial prediction device comprises a calculation module able to determine the motion vector of the current image partition from a function of at least one reference motion vector which belongs to a set of k reference motion vectors respectively associated with the neighboring k reference partitions.
• the present invention relates to a coding device for an image or a sequence of images generating a data stream comprising data representative of at least one image partition, such a device comprising means for spatial prediction of a motion vector of the image partition.
• Such a coding device is characterized in that the prediction means are contained in the above-mentioned spatial prediction device.
• the present invention relates to a device for decoding a data flow representative of an image or a sequence of images, said flow comprising data representative of at least one image partition.
• a device for decoding a data flow representative of an image or a sequence of images comprising data representative of at least one image partition.
• a device comprising means for spatial prediction of a motion vector of the image partition.
• Such a decoding device is characterized in that the prediction means are contained in the aforementioned spatial prediction device.
• the invention also relates to a computer program comprising instructions for implementing one of the methods according to the invention, when it is executed on a computer.
• the coding method, the decoding method, the spatial prediction device, the coding device and the decoding device have at least the same advantages as those conferred by the spatial prediction method according to the present invention.
• the invention is also applicable to a temporal prediction of the motion vector.
• the present invention relates to a method for temporally predicting a motion vector of a partition of a current image, the motion vector pointing to a reference image area that has the same shape. that the current partition and which belongs to a reference image different from the current image and having been cut beforehand, at the end of an encoding then a decoding, into a plurality of n partitions.
• the motion vector of the current image partition is determined from a function d at least one reference motion vector belonging to a set of k reference motion vectors respectively associated with k partitions of the plurality of n reference partitions, the k partitions being close to said reference image zone.
• the temporal prediction according to the invention is adaptable to any type of calculation method of the predicted motion vector of the current partition, such as in particular that conforming to the H264 / AVC standard and that described in the aforementioned IEEE publication.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select very precise reference motion vectors from a selection criterion which is here based on the proximity between the reference image area on the one hand and on the other hand the ( s) reference partition (s) closest to them.
• determining the motion vector of the reference image area comprises the steps of:
• the determination of the motion vector of the current partition comprises the steps of:
• Such arrangements thus make it possible to select even more finely and more precisely one or more reference motion vectors from a selection criterion which is here based not only on the proximity between the reference image area and the ( s) nearest reference partition (s), but also a comparison of the distance between the center of the reference image area on the one hand and the center of the reference image area on the other hand; of each of the neighboring reference partitions.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select, in a targeted manner, one or more specific reference movement vectors from a choice criterion which is here based on the existence of a discontinuity in previously coded macroblocks, then decoded.
• the accuracy of the temporal prediction is thus improved because obtained from reference motion vectors which are supposed to be more accurate in an area of the image which contains discontinuities than in a zone of the homogeneous image.
• determining the motion vector of the current partition comprises the steps of:
• Such an arrangement thus makes it possible to select very precise reference motion vectors from a selection criterion which is here based on the proximity between the reference image area on the one hand and on the other hand the ( s) reference partition (s) closest to them.
• Such a selection is further refined by determining the longest edge of the reference image area which delimits the latter from at least one neighboring reference partition.
• the selection step is followed by a step of calculating the average of the kbd reference movement vectors respectively corresponding to the k bd selected reference partitions.
• the selection step consists in choosing, in the subset of k bd reference partitions having said longest edge in common, the reference partition whose edge portion in common is the longest, then the reference movement vector corresponding to the selected reference partition.
• determining the motion vector of the current partition comprises the steps of: calculating, on the one hand, an average pixel value of the reference image area, and, on the other hand, a pixel value of the same type for each of the k neighboring reference partitions,
• Such an arrangement thus makes it possible to select a very precise reference motion vector from a certain characteristic which is here based on the identification of a particular pixellic component, such as for example the average luminance of a pattern, of a color, an outline, etc. in a reference partition adjacent to the reference image area.
• the present invention relates to a method for coding an image or a sequence of images generating a data stream comprising data representative of at least one image partition, such a method comprising a step temporal prediction of a motion vector of the image partition.
• Such a coding method is characterized in that the temporal prediction is performed in accordance with the aforesaid temporal prediction method.
• the present invention relates to a method for decoding a data flow representative of an image or a sequence of images, said flow comprising data representative of at least one image partition, a such method comprising a step of temporally predicting a motion vector of the image partition.
• Such a method is characterized in that the temporal prediction is performed in accordance with the aforesaid temporal prediction method.
• the present invention relates to a device for temporal prediction of a motion vector of a partition of a current image, the motion vector pointing to a reference image area that has the same shape as the current partition and that belongs to a reference image different from the current image and that has been previously cut out, resulting from an encoding then a decoding, into a plurality of n partitions.
• the temporal prediction device comprises a calculation module able to determine the movement vector of the partition current image from a function of at least one reference motion vector belonging to a set of k reference motion vectors respectively associated with k partitions of the plurality of n reference partitions, the k partitions being close to the reference image area.
• the present invention relates to a device for encoding an image or a sequence of images generating a data stream comprising data representative of at least one image partition, such a device comprising means for temporally predicting a motion vector of the image partition.
• Such a coding device is characterized in that the prediction means are contained in the aforementioned temporal prediction device.
• the present invention relates to a device for decoding a data flow representative of an image or a sequence of images, said flow comprising data representative of at least one image partition.
• a device for decoding a data flow representative of an image or a sequence of images comprising data representative of at least one image partition.
• a device comprising means for temporal prediction of a motion vector of the image partition.
• Such a decoding device is characterized in that the prediction means are contained in the aforementioned temporal prediction device.
• the invention also relates to a computer program comprising instructions for implementing one of the methods according to the invention, when it is executed on a computer.
• the coding method, the decoding method, the temporal prediction device, the coding device and the decoding device have at least the same advantages as those conferred by the temporal prediction method according to the present invention.
• FIG. 1A represents an example of temporal prediction of the prior art which exploits the temporal correlations between a current macroblock to be coded of an image N and a reference macroblock of a preceding image N-1, the reference macroblock having a square shape and being disposed within a neighboring reference macroblock,
• FIG. 1B represents a macroblock cut according to different types of partitions of the prior art
• FIG. 1C represents an example of temporal prediction of the prior art that exploits the temporal correlations between a current macroblock to be encoded of an N-picture and a reference macroblock of a previous N-1 picture, the current macroblock being cut off according to several partitions of arbitrary form, the motion vector of one of these partitions pointing to a reference image area which is arranged inside a neighboring reference macroblock,
• FIG. 1D represents an exemplary spatial prediction of the prior art that exploits the spatial correlations between a current macroblock to be encoded of an N image and reference partitions that are close to this image
• FIG. 2 represents the steps of the coding method according to the invention
• FIG. 3 represents an embodiment of a coding device according to the invention
• FIGS. 4A and 4B represent a first method for predicting a motion vector according to the invention, according to the two types of prediction, respectively spatial and temporal,
• FIGS. 5A and 5B show a second method for predicting a motion vector according to the invention, according to the two types of prediction, respectively spatial and temporal,
• FIGS. 6A and 6B show a third method for predicting a motion vector according to the invention, according to the two types of prediction, respectively spatial and temporal,
• FIGS. 7A and 7B represent a fourth method for predicting a motion vector according to the invention, according to the two types of prediction, respectively spatial and temporal,
• FIGS. 8A and 8B represent a fifth method for predicting a motion vector according to the invention, according to the two types of prediction, respectively spatial and temporal,
• FIG. 9 represents a decoding device according to the invention.
• FIG. 10 represents steps of the decoding method according to the invention. Detailed description of an embodiment
• the coding method according to the invention is used to code in Inter a sequence of images according to a bit stream close to that obtained by a coding according to the H.264 / MPEG-4 AVC standard.
• the coding method according to the invention is for example implemented in a software or hardware way by modifications of an encoder initially compliant with the H.264 / MPEG-4 AVC standard.
• the coding method according to the invention is represented in the form of an algorithm comprising steps CO to C6, represented in FIG. 2.
• the coding method according to the invention is implemented in a coding device CO represented in FIG.
• the first step CO is the selection, for a macroblock belonging to an image of the sequence of images to be encoded, denoted IN in FIG. 3, of a particular partitioning associated with this macroblock.
• step CO may be optional, the prediction of the motion vector of the current macroblock being able to be performed by considering the latter in its entirety, that is to say as a single partition.
• a macroblock MB N for example of size 4x4, and belonging to the image I N , is applied as input to a partitioning selection module SP shown in FIG. 3.
• This partitioning module SP uses for example a choice method by exhaustive competition or even a method of choice using an algorithm with a priori. Such methods are well known to those skilled in the art (see: GJ Sullivan and T. Wiegand, "Spleen-distortion optimization for video compression", IEEE Signal Proc.Mr., pp.74-90, 1998 / Elles will not be described further.
• the different types of possible partitioning algorithms are grouped together in a BD database of the CO encoder. They make it possible to obtain a division of the current macroblock into a plurality of partitions, either of rectangular or square shape, or of other geometrical shapes, such as, for example, substantially linear shapes, or of quite arbitrary shape.
• the next step C1 represented in FIG. 2 is the division of the macroblock MB N into a number of partitions to be predicted.
• the macroblock MBN is divided into, for example, four smaller partitions P1, P2, P3 and P4 of square shape.
• Such clipping is performed by a PMBCO macroblock partitioning module shown in Figure 3 which uses a conventional partitioning algorithm.
• the partitioning module PMBCO transmits the macroblock MB N which has just been partitioned to a prediction module PREDCO represented in FIG. 3.
• such a prediction module PREDCO is intended to predict the partitioned MBN current macroblock with respect to a plurality of n reference partitions belonging either to a previous IN- I image which has been previously coded and then decoded, or to the current image IN-
• the reference partitions of the previous IN- I image are coded in accordance with the H.264 / MPEG-4AVC standard, that is to say that they undergo, in a manner known per se, :
• PREDCO includes:
• a PMB partitioning module for cutting the reference macroblocks of the IN- I image according to a plurality of n reference partitions r'1, r'2, ..., r'n, a calculation module CAL intended to predict each motion vector MVpI, MVp2,..., MVpp which are respectively associated with the partitions P1, P2,..., Pp of the current macroblock MBN-
• the calculation module CAL predicts the motion vector of the current image partition from a function of at least one reference motion vector which belongs to a set of k reference motion vectors respectively associated with the k neighboring reference partitions.
• the partitioning module PMB of FIG. 3 cuts the reference image IN- I according to n reference partitions r'1, r'2 r'n.
• the calculation module CAL of FIG. 3 calculates, for each current partition, the predicted motion vector which is respectively associated with it, according to the various methods according to the invention described hereinabove. below.
• the module CAL dilates at least one pixel of the area d reference picture p'1.
• Such an expansion consists for example in a propagation of a mathematical operator of morphological dilation well known to those skilled in the art. In the example shown, the expansion performed is a single pixel and is represented by the hatched area in Figure 4A.
• the calculation module CAL selects a subset of k c neighboring reference partitions that overlap with the expanded reference picture area p'lext. In the example shown, this is k c neighboring reference partitions R'1, R'2, R "3, R'4, R'5 and R'6.
• the calculation module CAL determines the vector predicted movement MVpI of the current partition P1 as a function of the reference movement vectors MVr'1, MVr'2, MVr'3, MVr'4, MVr'5 and MVr'6 associated respectively with the six overlapping reference partitions r'1, r'2, r'3, r'4, r'5 and r'6 represented in FIG 4A
• Such a determination consists for example in calculating the median of the reference motion vectors MVr 1 I, MVr'2, MVr ' 3, MVr'4, MVr'5, MVr'6 according to the equation below:
• MVp 1 Moy (MVr '1, MVr'2, MVr'3, MVr'4, MVr'5, MVr'6)
• the module CAL determines an average of the reference movement vectors MVr' 1, MVr'2, MVr'3, MVr'4, MVr'5, MVr'6 which is weighted by the number of common pixels between the expanded reference image area p'l ext and each of the reference partitions r 1, r'2, r'3, r'4, r'5, r'6 overlapping the latter.
• Such a determination amounts to calculating the predicted motion vector MVpI according to the equation below:
• MVpl v 2Jfp -l1ext, -nr ⁇ kc l ⁇ ' mvr' * ⁇ plext nr VD with:
• MVpI ⁇ . (9r 'l + 4r' 3 -f4r '5 + r' 2 + r 4 + r 6).
• the first method and its variant which has just been described above can be applied to the spatial prediction of an MVpI motion vector as represented in FIG. 4B.
• the module CAL dilates at least one pixel of the current partition P1 as described above. with reference to FIG. 4A, so as to obtain a dilated current partition P1ext.
• the expansion performed is of a single pixel and is represented by the hatched area in FIG. 4B.
• the calculation module CAL determines the predicted motion vector MVpI of the current partition P1 as a function of the reference motion vectors MVM, MVr2, MVr3, MVr4, MVr5 and MVr6 associated respectively with the six overlapping reference partitions pr1, pr2, pr3, pr4 , pr5 and pr6 shown in Figure 4B. Such a determination consists for example in calculating the median of the reference movement vectors MVM, MVr2, MVr3, MVr4, MVr5, MVr6 according to the equation below:
• MVp1 Avg (MVM, MVr2, MVr3, MVr4, MVr5, MVr6)
• the module CAL determines an average of the reference movement vectors MVM, MVr2, MVr3, MVr4, MVr5, MVr6 which is weighted by the number of common pixels between the expanded current partition P1ext and each of the reference partitions pr1, pr2, pr3, pr4, pr5, pr6 overlapping the latter.
• the MB macroblock N is cut into for example three partitions P1, P2 and P3 of arbitrary form.
• the reference image I N- i is in turn cut out according to n reference partitions r'1, r "2, ..., r'n.
• the module CAL determines the center of the reference image area p'1, denoted by CTp'1 in FIG.
• CTr'1, CTr "2, CTr'3, CTr'4, CTr'5 of each of the neighboring reference partitions r'1, r'2, r0, r'4, r'5 CAL calculates then the distance which separates the center CTp'1 from the reference image zone p'1 with respectively each of the centers CTr'1, CTr'2, CTr'3, CTr'4, CTr'5.
• CTr'1, CTr ⁇ , CTr'3, CTr'4, CTr'5 are calculated by means of an algorithm which minimizes the sum of the distances with respect to all the points of the reference image area P ' 1, and parts r'1, r'2, r'3, r'4, r'5 respectively.
• the module CAL then selects, among the neighboring reference partitions r'1, r'2, r'3, r'4, r'5, the one whose center is situated at the lowest calculated distance of the zone of reference picture p'1. In the example shown, it is the reference partition r'1. Finally, the module CAL calculates the predicted motion vector MVpI as being equal to the reference movement vector MVr '1 corresponding to the selected reference partition r'1.
• the module CAL selects, from among all the neighboring reference partitions r'1 to r'5, those which are closest to the reference image area p'1. In the example shown, these are the reference partitions r'1, r'2 and r'3. The module CAL then determines the center CTp'1 of the reference image area p'1, and secondly, the corresponding center CTr'1, CTr'2, CTr'3 of each of the reference partitions r ' 1, r'2, r'3 the closest. The module CAL then calculates the distance that separates the center CTp '1 from the image area p'1 with respectively each of the centers CTr'1, CTr ⁇ , CTr "3. The module CAL calculates the predicted motion vector MVpI as being equal to the average of the reference motion vectors MVr 1 I, MVr'2, MVr 3, said average being weighted by the three respective distances calculated.
• the second method and its variant which has just been described above can be applied to the spatial prediction of an MVpI motion vector as represented in FIG. 5B.
• the module CAL determines the center of the current partition P1, noted CTpI in FIG. 5B, and on the other hand, the corresponding center CTrI, CTr2, CTr3, CTr4, CTr5 of each of the neighboring reference partitions pr1, pr2, pr3, pr4, pr5.
• the module CAL calculates then the distance which separates the center CTpI of the current partition P1 with respectively each centers CTrI, CTr2, CTr3, CTr4, CTr5.
• the module CAL then selects, among the neighboring reference partitions pr1, pr2, pr3, pr4, pr5, the one whose center is located at the lowest calculated distance of the current partition P1. In the example shown, it is the reference partition pr1. Finally, the module CAL calculates the predicted motion vector MVpI as being equal to the reference movement vector MVrI corresponding to the selected reference partition pr1.
• MVpI M VM.
• the module CAL selects, from among all the neighboring reference partitions pr1 to pr5, those which are closest to the current partition P1. In the example shown, these are the reference partitions pr1, pr2 and pr3.
• the module CAL determines the center CTpI of the current partition P1, and on the other hand, the corresponding center CTrI, CTr2, CTr3 of each of the reference partitions pr1, pr2, pr3.
• the module CAL then calculates the distance that separates the center CTpI from the current partition P ⁇ with respectively each of the centers CTrI, CTr2, CTr3.
• the CAL calculates the predicted motion vector MVpI as equal to the average of the reference movement vectors MVM, MVr2, MVr3, said average being weighted by the three respective distances calculated.
• the MB macroblock N is cut into for example three partitions P1,
• the reference image l N -i is in turn cut out according to n reference partitions r'1, r'2, ..., r'n.
• the module CAL proceeds to extend an edge bd'1 of the reference image area p'1 on at least some of the partitions of reference neighbors above.
• the extension is represented by a dashed line in Figure 6A.
• the CAL module selects, from among the neighboring reference partitions, the reference partitions that are located on the same side of the extended edge. In the example shown, these are the reference partitions r'1, r'2, r'3.
• the module CAL then calculates the predicted motion vector MVpI as being equal to the average of the reference movement vectors MVr'1, MVr'2, MVr'3 which are associated respectively with the selected reference partitions r'1, r'2, r '3.
• the third method that has just been described above can be applied to the spatial prediction of an MVpI motion vector as represented in FIG. 6B.
• the module CAL proceeds to extend an edge bd1 of the current partition P1 on at least some of the aforementioned neighboring reference partitions.
• the extension is represented by a dashed line in Figure 6B.
• the CAL module selects, from among the neighboring reference partitions, the reference partitions that are located on the same side of the extended edge. In the example shown, these are the reference partitions pr1, pr2, pr3.
• the module CAL calculates the predicted motion vector MVpI as being equal to the average of the reference movement vectors MVrI, MVr2, MVr3 which are respectively associated with the selected reference partitions pr1, pr2, pr3.
• the macroblock MBN is divided into two arbitrary-shaped partitions P1 and P2, for example.
• the reference image l N -i is in turn cut out according to n reference partitions r'1, r'2, ..., r'n.
• the module CAL determines the longest edge of the reference image area p'1. Such an edge is denoted bdl'1 in FIG. 7A.
• the module CAL selects, among the neighboring reference partitions r'1 to r'5, the reference partitions having the edge bdl'1 in common. In the example shown, these are the reference partitions r'3 and r'4.
• the module CAL calculates the predicted motion vector MVpI as being equal to the average of the reference movement vectors MVr'3 and MVr'4 which are associated respectively with the selected reference partitions r'3 and r '4.
• the module CAL weight said calculated average by the length of the edge of the reference partition r'3 which is in common with the reference image area p'1 and by the length of the edge of the partition of reference r'4 which is in common with the reference image area p'1.
• the module CAL selects, from among the two reference partitions r'3 and r'4, the one whose edge in common with the reference image zone p'1 is the longest. In the example shown, it is the reference partition r'3.
• the module CAL calculates the predicted motion vector MVpI as being equal to the reference movement vector MVr 1 S corresponding to the selected reference partition r'3.
• MVp1 MVr'3.
• the fourth method and its three variants which have just been described above can be applied to the spatial prediction of an MVpI motion vector as represented in FIG. 7B.
• the module CAL determines the longest edge of the current partition P1.
• Such an edge is denoted bdh in FIG. 7B.
• the module CAL selects, among the neighboring reference partitions pr1 to pr5, the reference partitions having the edge bdh in common. In the example shown, these are the reference partitions pr3 and pr4.
• the module CAL the module CAL:
• the MB macroblock N is cut into for example two equal partitions P1 and P2 of rectangular shape.
• the reference image I N- i is in turn cut according to n reference partitions r'1, r * 2 r'n.
• the module CAL calculates the average luminance of the reference image area p'1 and each of the neighboring reference partitions r'1 to r'5. The latter then selects, from among the neighboring reference partitions r'1 to r'5, the one whose calculated average luminance is closest to that of the reference image area p'1.
• the reference partition r'2 which has a pattern C'2 similar to that of the reference image area p'1.
• the module CAL then calculates the predicted motion vector MVpI as being equal to the reference movement vector MVr'2 corresponding to the selected reference partition r'2.
• the fifth method that has just been described above can be applied to the spatial prediction of an MVpI motion vector as represented in FIG. 8B.
• the CAL module calculates the average luminance of the current partition P1 and of each of the neighboring reference partitions pr1 to pr5. The latter then selects, from among the neighboring reference partitions pr1 to pr5, the one whose calculated average luminance is closest to that of the current partition P1. In the example shown, this is the reference partition pr2 which has a Cr2 pattern similar to that of the current partition P1.
• the module CAL then calculates the predicted motion vector MVpI as being equal to the reference movement vector MVr2 corresponding to the selected reference partition pr2.
• the prediction calculation module PREDCO then delivers a first predicted vector MVpI which, in the case where this it is retained by the CO encoder as the optimal motion vector type, is immediately coded by the TQCO transform and quantization module, and then decoded by the inverse transform and quantization module TQICO, which are represented in FIG.
• step C4 is then reiterated so as to predict the other motion vectors MVp2 to MVpp which are respectively associated with the partitions P2 to Pp of the current macroblock MB N -
• a decision module DCNCO traverses the partitioned macroblocks of the image IN and FIG. chooses, in this step C5, the prediction mode used to code each of these macroblocks.
• the decision module DCNCO chooses the optimal prediction according to a distortion flow criterion well known to those skilled in the art.
• each predicted macroblock is coded, during a step C6, as in the H.264 / MPEG-4 AVC standard.
• the residual coefficients if they exist, corresponding to the blocks of the image I N , are sent to the TQCO module for transforming and quantizing, to undergo discrete cosine transforms and then quantization.
• the macroblock slices with these quantized coefficients are then transmitted to an EC coding module entropic shown, to produce, with the other images of the video sequence already coded in the same way as the IN image, a video stream F, binary coded according to the invention.
• the bit stream F thus coded is transmitted by a communication network to a remote terminal.
• This comprises a decoder DO according to the invention, represented in FIG. 9.
• the bit stream F is first sent to an entropy decoding module DE, decoding in the opposite direction to that performed by the entropy coding module CE represented in FIG. 3. Then, for each image macroblock to be reconstructed, the coefficients decoded by the DE module are sent to a QTIDO module for inverse quantization and inverse transform.
• An image reconstruction module R 1 then receives decoded data corresponding to the data produced by the DCNCO module (FIG. 3) at coding step C5 according to the invention, with transmission errors close to it.
• the module R1 implements steps D0 to D6 of the decoding method according to the invention, as represented in FIG.
• Such a decoding method according to the invention is also implemented in a software or hardware way by modifying a decoder initially compliant with the H.264 / MPEG-4 AVC standard.
• the first step DO is the decoding of coded data structures in a slice of a current macroblock of the image I N to be decoded.
• the reconstruction module R1 determines from the data of said macroblock slice:
• Inter 4x4 in the embodiment described,
• the next step D1 shown in FIG. 10 is the division of the current macroblock to be decoded, in accordance with the partitioning determined in step D0.
• a macroblock partitioning module PMBDO which in all respects resembles that shown in FIG. 3, cuts the macroblock into a plurality of p partitions, P1 to Pp.
• the partitioning module PMBDO transmits the current macroblock to be decoded, which has just been partitioned, to a prediction module PREDDO represented in FIG. 9, which is in all respects similar to the PREDCO prediction module of the CO encoder of FIG. 3, and which, for this reason, will not be described again in detail.
• the prediction module PREDDO of FIG. 9 performs the same algorithm as that carried out by the prediction module PREDCO of the aforementioned CO encoder, so as to obtain a current macroblock whose vectors Associated movements have been predicted according to one or other of the five methods described above.
• a decision module DCNDO chooses the optimal prediction according to a distortion flow criterion well known to those skilled in the art.
• Each predicted macroblock is then decoded, during a step D6, as in the H.264 / MPEG-AVC standard.
• the image reconstruction module R 1 delivers, at the output of the decoder DO, an IDN image corresponding to the decoding of the image I N.

Landscapes

• Engineering & Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Compression Or Coding Systems Of Tv Signals (AREA)
• Compression, Expansion, Code Conversion, And Decoders (AREA)

Priority Applications (2)

Applications Claiming Priority (2)

Related Child Applications (2)

Publications (1)

Family

ID=41445499

Family Applications (3)

Family Applications Before (2)

Country Status (4)

Families Citing this family (15)

Family Cites Families (12)

• 2010
  • 2010-06-25 CN CN201080038820.XA patent/CN102498718B/zh active Active
  • 2010-06-25 WO PCT/FR2010/051306 patent/WO2011001078A1/fr active Application Filing
  • 2010-06-25 EP EP20156424.2A patent/EP3678374A1/de active Pending
  • 2010-06-25 EP EP19215772.5A patent/EP3651466A1/de active Pending
  • 2010-06-25 US US13/381,072 patent/US10051283B2/en active Active
  • 2010-06-25 EP EP10745326A patent/EP2449784A1/de not_active Ceased

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events