FR2766033A1 - Compressed data coding-decoding method - Google Patents

Abstract

The coding-decoding method operates on a set of data representing physical quantities, the set being comprised of sub-sets fn. For each sub-set it is divided into blocks of data to be coded, a reference set is divided into source blocks and then, for each block of data to be coded, a predetermined number of source blocks in the reference sub-set are selected. The method then determines a transformation between the source blocks selected and the block to be coded.

Description

 The present invention relates, in general, to a method and a device for encoding data, as well as to a method and a device for decoding data. The coding and decoding of data aims in particular at compressing and decompressing sets of data.

 A compressed data set can be memorized by occupying a reduced memory space compared to the initial uncompressed set, or can be transmitted more quickly, or with a lower bit rate, than the initial uncompressed set.

 These data are a priori arbitrary and represent physical quantities. In preferred applications, we consider data sets which are sequences or sequences of images, sound or other mono or multidimensional signals. In the following, we will consider more particularly the coding and decoding of image sequences, since the images are known on the one hand to occupy a large memory space and on the other hand to require a high bit rate, or a time of important transmission, when one wishes to transmit them.

Many methods of coding image sequences currently exist. A first type of process is coding known as MPEG (Motion
Picture Expert Group) which generally consists of coding images in relation to one or more other images selected in the sequence, called reference images, and periodically an image in relation to itself.

 An image coded with respect to itself, known as an "intra" image, is divided into blocks, which are coded by discrete cosine transformation (DCT).

These images are necessary for refreshing and allow random access in the sequence; however, the compression rate of these images is low.

 An image coded with respect to one or more other images is also divided into blocks. Each block is coded by an estimation and motion compensation mechanism with respect to a block which must be sought in the reference images, then the error is coded by discrete transformation into cosine. The search for the "best" block for coding the current block is generally carried out in a window of the reference image, to reduce the search time which could otherwise be too long. However, this type of coding remains sluggish.

 In addition, this type of coding is not suitable for low bit rates, of the order of a few tens of kilobits, but is limited to bit rates generally greater than 1.5 megabits per second.

 A second type of image sequence coding consists in coding a current image of the sequence with respect to the previous image. The current image is divided into blocks. For each block to be coded, a single block is sought in a dictionary consisting of the blocks of the previous image, having undergone transformations such as rotation, or symmetry. A linear transformation is then calculated between the current block and the reference block.

 These two types of coding method have in common the search for, for each block to be coded, a reference block in a set of blocks, which is relatively long.

 In addition, the second type of coding involves decoding by iteration of each coded block, which again requires a relatively long time.

 The present invention aims to remedy the drawbacks of the prior art, by providing a method and a device for coding data, as well as the corresponding method and device for decoding, having a high data compression rate, while being fast at coding. and decoding.

 The invention relates to the field of lossy coding and decoding of block digital image sequences.

 The invention is based on the principle of coding each image of the sequence as a function of at least one reference image.

The invention provides a method of coding a set of data representative of physical quantities, said set being composed of subsets,
characterized in that it comprises, for each of the subsets, the steps of:
- division of the subset into data blocks to be coded,
- division of at least one reference subset into source blocks,
then, for each of the data blocks to be coded, the steps of:
selection of a predetermined number of source blocks in said at least one reference subset,
- determination of a transformation between the selected source blocks and the block to be coded.

Correlatively, the invention proposes a device for coding a set of data representative of physical quantities, said set being composed of subsets,
characterized in that it comprises:
means for dividing the subsets into data blocks to be coded,
means for dividing at least one reference subset into source blocks, for each of the subsets,
means for selecting a predetermined number of source blocks in said at least one reference subset, for each of the data blocks to be coded from each of the subsets, and
means for determining a transformation between the selected source blocks and the block to be coded.

 The invention makes it possible to obtain a high compression rate of the data, which can then be transmitted at low speed, of the order of a few tens of kilobits per second, while offering a good quality of data reconstruction.

 The invention allows rapid compression and decompression of data.

 The invention applies equally well to the case where both rapid compression and decompression of image sequence are desired, as well as to the case where only rapid decompression is desired.

 According to a preferred embodiment, the method further comprises the step of searching if the block to be coded can be coded by block matching, and of only searching for a transformation if a matching criterion is not satisfied. Block mapping coding requires few bits to store and / or transmit the coded block.

 The correspondence criterion comprises a measurement of a first distance between the block to be coded and the source block, which constitutes a simple means of implementing this aspect of the invention.

In another preferred embodiment of the invention, the method comprises, for each of the subsets, the additional steps of:
- division of the sub-assembly into zones,
determination of the type of each zone, from a first type of zone and a second type of zone, according to a zone criterion,
- division of each second type area into data blocks to be coded. The second type areas are then coded in blocks, as explained above.

 The method comprises the calculation of a second distance between each zone and a corresponding zone in an adjacent subset, the zone criterion consists in comparing this distance with a threshold, and, if this distance is less than the threshold, the zone considered is first type, and second type otherwise. Thus, the zones of the first type are zones which remain unchanged from one subset to another. These areas are coded by another coding method, or are copied from one subset to another, without being coded. The zones which remain unchanged in several subsets are coded only once, for the first of the subsets considered, then are copied for the following subsets, which is rapid and requires few bits to store and / or transmit these zones. Decoding is also fast.

 In the case where the data set to be coded is a sequence of digital images, the areas of the first type are coded by a method of coding a fixed image.

 In another aspect, the transformation is determined so as to minimize a third distance between the block to be coded and its approximation calculated by applying the transformation to the selected source blocks. The calculation of the third distance comprises the calculation of a difference, for example the mean square error, between the values of the data of the block to be coded and the values of its approximation calculated by applying the transformation to the source blocks.

 According to preferred characteristics, the transformation is a multilinear approximation, or the transformation is a multilinear approximation combined with a geometric transformation, or else the transformation is a polynomial on the values of the coefficients of the source blocks. The type of transformation is chosen as a function of the coding precision, the coding speed and the compression rate desired.

According to preferred characteristics:
- for the first subset to be coded, said at least one reference subset is a predetermined subset,
- Said at least one reference subset is a subset of the set, which has preferably been previously coded and then decoded.

According to other preferred characteristics:
- at least two reference subsets are considered, at least one of which is located among the subsets preceding the subset being coded and at least one second of which is located among the subsets following the subset being coded ,
- the source blocks are selected from a part of the reference subset which depends on the part in which the block to be coded is located in the subset being coded,
- the source blocks have a size which is determined as a function of the size of the block to be coded,
the source blocks have a size multiple of a factor F2 from that of the block to be coded and are subsampled by the factor F, where F is an integer, or the source blocks have the same size as the block to be coded, which is easier to do.

According to other preferred characteristics:
- the block to be coded is of predetermined size, or
- the block to be coded is of variable size, according to a size criterion.

 In the latter case, the method comprises the calculation of a fourth distance between the block to be coded and the approximation calculated by applying the transformation to the source blocks, the size criterion consists in comparing the fourth distance to a threshold, and if the fourth distance is greater than the threshold, and if a minimum block size has not been reached, the block being coded is divided into blocks of smaller size.

 The coding device comprises means adapted to implement the characteristics described above.

These means are incorporated into:
- a controller,
a read-only memory comprising a program for coding each of the data blocks, and
- a random access memory comprising registers adapted to record variables modified during the execution of said program.

The invention also relates to a method for decoding a set of data representative of physical quantities, the set being composed of subsets which each include coded representations of data blocks, each of the blocks being coded by means of a transformation between source blocks selected in at least one reference subset and the block,
characterized in that it comprises, for each of the blocks of each of the subsets to be decoded, the step of applying the transformation to the source blocks to decode the block.

In a preferred embodiment, the decoding method comprises, for each of the blocks of each of the subsets to be decoded, the steps of:
- research if the block is coded by matching of blocks,
and, in the event of a negative response, the block then being coded by means of a transformation between source blocks selected in at least one reference subset and the block:
- application of the transformation to the source blocks to decode the block.

 In the event of a positive response to the search step, the block is decoded by copying a correspondence block determined by the matching of blocks. The decoding of the block considered is therefore very simple and rapid.

 According to a preferred characteristic, the decoding method comprises reading the size and the position of the blocks to be decoded, when the blocks are of variable size.

In another aspect, the invention relates to a method of decoding a set of data representative of physical quantities,
The assembly being composed of subsets which each comprise coded representations of zones belonging to a first or a second type of zone, the coded representations of the second type of zone being produced by data blocks,
characterized in that it comprises, for each of the blocks of the second type zones of each of the subsets to be decoded, the step of applying the transformation to the source blocks to decode the block.

The invention also relates to a decoding device comprising means suitable for implementing the characteristics described above. These means are incorporated into:
- a controller,
a read only memory comprising a program for decoding each of the data blocks, and
- a random access memory comprising registers adapted to record variables modified during the execution of said program.

The characteristics and advantages of the present invention will appear more clearly on reading a preferred embodiment illustrated by the attached drawings, in which:
FIG. 1 schematically represents a data coding device according to the invention,
FIG. 2 schematically represents a device for decoding data according to the invention,
FIG. 3 represents an embodiment of the data coding device according to the invention,
FIG. 4 represents an embodiment of the data decoding device according to the invention,
FIG. 5 is a data coding algorithm according to a first embodiment of the invention,
FIG. 6 is an image divided into blocks during its coding according to the invention,
FIG. 7 is an image divided into blocks during its coding according to the invention,
- Figures 8a to 8f show blocks to be coded and associated source blocks
FIG. 9 is a data decoding algorithm according to the first embodiment of the invention,
FIG. 10 is a data coding algorithm according to a second embodiment of the invention,
FIG. 11 is an image divided into blocks during its coding according to the invention,
FIG. 12 is a cutting tree associated with the image of FIG. 11,
- Figure 13 is a data decoding algorithm according to the second embodiment of the invention
FIG. 14 is a data coding algorithm according to a third embodiment of the invention,
FIG. 15 is a data decoding algorithm according to the third embodiment of the invention,
FIG. 16 is a data coding algorithm according to a fourth embodiment of the invention,
FIG. 17 is an image divided into zones during its coding according to the fourth embodiment of the invention, and
- Figure 18 is a data decoding algorithm according to the fourth embodiment of the invention.

 According to an embodiment chosen and shown in FIG. 1, a data coding device 2 according to the invention comprises an input 22 to which is connected a source 1 of non-coded data. The source 1 comprises for example a memory means, such as random access memory, hard disk, floppy disk, compact disc, for memorizing non-coded data, this memory means being associated with a suitable reading means for reading the data there. Means for storing the data in the memory means can also be provided.

 The source 1 may for example be a video camera in the case of an image signal or a microphone in the case of a sound signal.

 Means 3 users of coded data are connected at output 27 of the coding device 2.

 The user means 3 comprise for example means for storing coded data, and / or means for transmitting the coded data.

 FIG. 2 represents a device 5 for decoding data coded by the device 2.

 Means 4 users of coded data are connected at the input 52 of the decoding device 5. The means 4 comprise for example coded data memory means, and / or means for receiving the coded data which are adapted to receive the coded data transmitted by the transmission means 3.

 Means 6 users of decoded data are connected to the output 57 of the decoding device 5. The user means 6 are for example means for viewing images, or means for restoring sounds, depending on the nature of the data processed.

 The coding device and the decoding device can be integrated into the same digital device, for example a digital camera provided with a display screen.

 An embodiment of the data coding device 2 is shown in FIG. 3.

The coding device 2 comprises a data and address bus 20 to which are connected:
an input buffer memory 21, also connected to input 22 of the device, for temporarily storing the data to be coded,
a read only memory (ROM) 23 in which is stored a coding algorithm which will be explained below,
a controller 24 for implementing the coding algorithm,
a random access memory (RAM) 25 for storing in registers parameters modified during the execution of the coding algorithm, and
- an output buffer memory 26, also connected to output 27 of the device.

 An embodiment of a data decoding device 5 is shown in FIG. 4.

the decoding device 5 comprises a data and address bus 50 to which are connected:
an input buffer memory 51, also connected to input 52 of the device, for temporarily storing the data to be coded,
a read only memory (ROM) 53 in which is stored a decoding algorithm which will be explained below,
a controller 54 for implementing the decoding algorithm,
a random access memory (RAM) 55 for storing in registers parameters modified during the execution of the decoding algorithm, and
- an output buffer memory 56, also connected to output 57 of the device.

 The operation of the coding 2 and decoding 5 devices will be detailed during the description of the various embodiments of the corresponding algorithms.

According to the embodiment chosen and represented in FIG. 5,
The invention applies to the coding of a set of data representative of physical quantities which is a sequence of digital images f1 to fN, where N is an integer.

 Any image fn in the sequence, with n between 1 and N, is a gray-level image, comprising for example 128 × 96 pixels at 256 levels.

 The algorithm represented in FIG. 5 comprises steps E1 to E14 which carry out the coding of the sequence of digital images. The processing is carried out image by image. Overall, an image of the sequence is coded as a function of a reference image.

 Step E1 is an initialization to which the parameter n is initialized to 1, so as to first consider the first image of the sequence.

 The next step E2 is a division of the image considered fn into a predetermined number M of blocks Anal1 to An, M of adjacent pixels. The number M of blocks can be predetermined, or can be a parameter set by the user of the algorithm. In this embodiment, the blocks A, 1 to An, M are adjacent, without overlap. In a variant not shown, the blocks can overlap, to limit the block effects on decoding, to the detriment of the compression rate.

 According to a first variant of this embodiment, shown in FIG. 6, the blocks all have the same predetermined size, and are for example square blocks of 32 × 32 pixels in the case of the 128 × 96 pixel image. The image division is uniform throughout the image. According to a second variant represented in FIG. 7, the size of the blocks, while being predetermined, depends on the location of the blocks in the image: in a determined area of the image, for example a central area capable of containing more details, the blocks are smaller than at the edges of the image.

This determined area will then be coded more finely than the rest of the image.

This variant is particularly advantageous for videophone, since in this case, the images mainly reproduce the faces of the users, who are generally located in the center of the images.

 Two different block sizes are shown in Figure 7 32x32 and 16x16 pixels. However, a higher number of sizes can be envisaged for specific applications, in particular if it is known that the quantity of information varies greatly from one zone to another of the images to be coded.

 In all cases, the division of the image results in a set of M blocks An.i to An, M, stored in memory 25. The blocks An 1 to An, M will be compressed, or coded, one at a time. following the others, in any order. This order is preferably predetermined and is then the order in which the blocks will then be decompressed, or decoded, which makes it unnecessary to locate the blocks by an index or by coordinates. As a variant, it is possible to locate the blocks and to associate the identification, index or coordinates, with the coded block, which makes it possible to decompress the blocks in any order, which may be different from the order used for compression.

 Step E2 is followed by step E3, which is a test on the value of the parameter n. If the parameter n is equal to 1, step E3 is followed by step E4 in which an additional image g0 is added to the sequence as a reference image for the first image f1 of the sequence. The additional image g0 is arbitrary, for example a uniformly gray image.

 If in step E3 the parameter n is greater than 1, step E3 is followed by step E5, just as step E4 is followed by step E5. The latter is an initialization of a parameter m to 1, to consider the first block A in image f ,.

 The next step E6 is the division of one or more reference images into blocks of adjacent pixels, stored in memory 25. The size of the blocks depends on the size of the blocks to be coded An, m. Preferably, the blocks of the reference images are F2 times larger than the blocks to be coded, where F is an integer for example equal to two, which makes it possible to exploit many details of the reference image. The blocks of the reference images are then sub-sampled by the factor F, to work on blocks having the same number of pixels as the blocks to be coded. According to a variant that is simpler to implement, the blocks of the reference images have the same size as the blocks to be coded.

Step E6 is followed by step E7 which is the selection of source blocks Dni, m, i to Dn1, m, K in one or more reference images relative to the block to be coded Anm of the image f, according to l one of the following variants
the source blocks are selected in a reference image which precedes the image being coded;
- the source blocks are selected from several reference images which precede the image being coded;
the source blocks are selected in a reference image which precedes, and in a reference image which follows, the image being coded;
the source blocks are selected from several preceding reference images, and / or from several reference images which follow,
The image being coded.

 In each of the preceding variants, the reference image or images are images of the sequence, or are images of the sequence which have been coded and then decoded according to the invention. This last case makes it possible to better restore the image during decoding which will be explained below. Indeed, only decoded images will be available during decoding; it is therefore preferable to also use decoded images as reference images for coding so as not to introduce an error.

 According to a preferred embodiment which is more particularly taken into account here, we consider a single reference image g1 which is the image preceding the image fn, previously coded and then decoded according to the invention.

 The number K of source blocks is predetermined, and is for example equal to 4. As a variant, the number K is a parameter for adjusting the algorithm. The source blocks are selected by their location in the reference image gn-1 as a function of the location of the block to be coded An, m in the image f ,. In general, the source blocks are selected in a region of the previous image which depends on the region of the block to be coded in the image considered. Preferably, the two regions are similar, that is to say that the superposition of the image under consideration and of the source image leads to partial overlapping of the source blocks by the block to be coded.

 As a variant, it is possible to select the source blocks in a "translated" region relative to the region of the block to be coded. This variant is advantageous in the case where it is known that the image to be coded represents a scene translated with respect to the reference image.

 Figures 8a to 8f show different examples of source block selection. In each of FIGS. 8a to 8f, the block to be coded An.m of the image considered fn has been superimposed on the source blocks Dn.i, mi to Dn1, m4 of the source image gon 1. In the six examples, we consider four source blocks of the same size as the block to be coded An, m.

 In FIG. 8a, the block to be coded An, m is in a central part of the image f,; the four source blocks are adjacent and the block to be coded is generally centered on the four selected source blocks.

 In FIGS. 8b to 8e, the block to be coded An, m is located on an edge of the image represented by a line in dotted lines. Two of the source blocks partially overlap the other two source blocks, so that the block to be coded covers part of each of the source blocks.

 In FIG. 8f, the block to be coded An, rn is in a corner of the image, and the four source blocks partially overlap, so that the block to be coded covers part of each of the source blocks.

 As a variant, when the block to be coded is on an edge or in a corner of the image, the four source blocks are adjacent, and two or three of the source blocks are truncated and completed either by a uniformly gray complement or by a complement determined by extension of the image, for example by symmetry or by periodization of the image.

 Step E7 is followed by step E8, at which a transformation between the source blocks and the block to be coded is calculated.

 According to a preferred embodiment, the transformation is a multilinear approximation, of the form: Tn.m = n, mk, bnm} making it possible to construct the block ARn, m = #k (an, m, k X Dn-1, m, k) + bn, m.

 In this expression, an mk is a scale function, here a constant function, or coefficient, for a given source block Dn1mk and of norm less than unity, and bn, m is a gray level shift function, here a constant function, or coefficient, for the block to be coded Anm. The function an, m, k applies to each of the pixels of the block Dn-1, m, k, which amounts here to multiplying all the pixels by the coefficient n, m, k.

 The functions, here the coefficients, an, m, k and bn, m are determined to minimize a distance d (An, m, ARn, m) between the block to be coded An, m and the reconstructed block ARnrn by applying the transformation Tn , m to the source blocks according to the formula given above.

According to an embodiment which is simple to implement, the functions an, m, k and bn, m are discretized and a limited number of possible values for the coefficients an, m, k is tested. For example, each
- the transformation combines a multilinear approximation calculated on the source blocks and geometric transformations, such as rotation, of the source blocks,
the transformation is of polynomial form, for example of order two or three, calculated on the values of the pixels of the source blocks. This type of transformation gives more precise results, while taking longer to calculate.

 The following step El 0 is a test to check if m is equal to M, that is to say if the block which has just been coded is the last of the current image. If at least one block remains to be coded in the image fn, the step ElO is followed by the step El 1 in which the parameter m is incremented by one, to consider the next block in the image fn. Step E11 is followed by step E7 previously described.

When the block which has just been coded is the last of the current image, step E10 is followed by step E12 which is a test to check whether the image fn which has just been compressed is the last of the sequence. If this is not the case, step E12 is followed by step E13 which is the construction of the decompressed image gon. The application of each of the transformations Tn, m previously determined to the source blocks reconstructs each of the blocks ARn, m, the whole of which forms the decoded image gn. The image gn is stored in memory 25 to be used for coding the following image fn + i (step
E6).

 Step E13 is followed by step E14 to increment the parameter n by one. Step E14 is followed by step E2 previously described.

In the case where n = N, that is to say for the last image of the sequence, it is not necessary here to reconstruct the image gN.

 When the response to step E11 is positive, all the images f1 to fN of the sequence have been compressed into compressed images h1 to hN.

 According to another embodiment, not shown, the coding method is modified in the following manner.

 To code an image of the sequence, we consider a reference image, chosen as described above.

 The source blocks are no longer selected for each of the blocks to be coded in the image; source blocks are selected in the reference image, once and for all for the image to be coded, and are then used to code all the blocks of the image to be coded. Coding is thus faster, since the selection of the source blocks is carried out only once per image; the distortions introduced by the coding will however be greater than with the embodiment described above.

 With reference to FIG. 9, the first embodiment of the algorithm for decoding the compressed sequence of digital images comprises steps E20 to E30. Step E20 is an initialization at which the parameter n is initialized at 1, to process the first compressed image h1 of the sequence.

In the next step E21, an additional image g0 is added to the sequence as a source image for the first compressed image h1 of the sequence. The additional image g0 is identical to that considered in step
E4 previously described, and is for example a uniformly gray image.

 The next step E22 is the initialization of the parameter m to 1, to consider the first block to be reconstructed. The blocks are preferably treated in the same order as in compression, which avoids having to memorize and process an index, or coordinates, to locate the blocks.

 The next step E23 is the division of the reference image g.1 into blocks of adjacent pixels, which are the source blocks stored in memory 55.

Step E23 is identical to step E6. The next step E24 is the selection of source blocks Dnî, mi to Dn-l, m, K in the reference image go 1, relative to the block to be reconstructed ARnm. The reference image is selected in the same way as for coding, with the same possible variants. According to the preferred embodiment, the reference image g1 is the decoded image preceding the image hn. The source blocks are selected in the same way as for coding (step E6), with the same possible variants.

Step E24 is followed by step E25 in which the transformation Tn, m defined by the functions an, m, k and bn, m is applied to the source blocks Dn 1 m 1 to Dn-l, m, K selected at the previous step E24, according to the formula:
ARn, m = Zk an, m, k X Dn-1, m, k) + bn, m.

 Step E25 results in the decoded block ARn, m stored in memory 55.

 The next step E26 is a test to determine whether the block which has just been decompressed is the last of the image being decompressed. If this is not the case, step E26 is followed by step E27 in which the parameter m is incremented by one unit to consider the next block. Step E27 is followed by step E24 previously described.

When all the blocks of the current image have been decompressed,
Step E26 is followed by step E28 in which the decompressed image gn is reconstructed and for example displayed on a screen, and / or stored. In general, Step E28 is the transfer of the image gn to the user means 6.

 The next step E29 is a test to determine whether the image which has just been decompressed is the last image of the sequence. If at least one image remains to be decompressed, step E29 is followed by step E30 in which the parameter n is incremented by one, to consider the next image in the sequence. Step E30 is followed by step E22 previously described.

 Another embodiment of the coding method according to the invention is represented in FIGS. 10, 11 and 12 in the form of an algorithm comprising steps E31 to E41 stored in memory 23 (FIG. 3).

 In this embodiment, the size of the blocks to be coded is determined adaptively, as a function of the quantity of information present in the blocks to be coded and as a function of the "movement" of the blocks to be coded with respect to the image of reference, that is to say modifications between the pixels of the block to be coded and the pixels of the source blocks of the reference image.

 A cutting tree, here a quaternary tree, is associated with each image to be coded. The quaternary tree represents the particular cutout of each image. Each compressed image hn has a set of transformations {Tn, m} and the associated quaternary tree.

 Step E31 is an initialization of a parameter n to 1, to first consider the first image f1 of the sequence.

 The next step E32 is the division of the current image fn into blocks of initial size. For example, as shown in Figure 11, The image fn has a size of 128x96 pixels, and it is divided into twelve blocks of size 32x32 pixels.

 Each block of initial size An to through An 12 is associated with a root of the quaternary tree represented in FIG. 12, that is to say with a node of the tree which has no associated data and which has not yet been subdivided. A marker is initialized to zero for each root.

 The next step E33 is the search for a transformation between a block An, m of the image to be coded fn and source blocks. Step E33 is similar to steps E3 to E8 previously detailed with reference to FIG. 5.

 Step E33 results in a transformation Tn, m determined for the block to be coded An, m. The different possible variants for the selection of the reference image and the source blocks, for the choice of the transformation and of the distance, previously exposed, are here applicable in the same way as for the first embodiment.

 The next step E34 is a test to determine whether the distance d (Anm, ARn, m) calculated in the previous step is less than a predetermined threshold S.

 If the answer is positive, this means that the calculated approximation makes it possible to reconstruct an ARnm block sufficiently close to the block to be coded An, m; Step E34 is followed by step E35 in which the transformation Tn, m is stored as a data block D associated with the node of the block Ans.

in the quaternary tree of the image fn. The marker remains at zero.

The reconstructed block ARn, m is calculated and stored to construct the image gn.

 Step E35 is followed by step E36 in which it is tested whether the block which has just been processed is the last one of the image being coded. If this is not the case, step E36 is followed by step E37 to consider another block of the image f i. Step E37 is followed by step E33 previously described.

When all the blocks of the image fn have been coded, step E36 is followed by step E38 in which it is tested if all the images of the sequence have been processed. If at least one image remains to be coded, step E38 is followed by step
E39 to consider another image of the sequence. Step E39 is followed by step E32.

If, at step E34, the distance d (An, m ARnm) calculated in the previous step E33 is greater than the threshold S, Step E33 is followed by step
E40 at which it is tested if the size of the block to be coded current Anm has reached a predetermined minimum value. If the answer is positive, step E40 is followed by step E35, that is to say that the determined transformation is stored, even if it does not satisfy the criterion of step E34, in the case where the size of the block to be coded current An, ma reaches the minimum value.

 If the response to the test in step E40 is negative, this step is followed by step E41 in which the current block to be coded An, m is divided into a predetermined number M1 of sub-blocks A1n.m to Ai name As shown in Figure 11, the division is for example carried out in four sub-blocks. In the quaternary tree, the marker of the node associated with the divided block An, m is set to the value 1, to indicate that the corresponding block has been divided. The division of the block results in four sub-blocks, whose nodes in the quaternary tree have markers initialized to zero. The order of the sub-blocks in the quaternary tree corresponds to their position in the divided block, and this order is the same for all the divided blocks: for example, lower right sub-block, lower left sub-block, top right block, top left sub-block. The position information of the sub-blocks is used during decoding.

 Step E41 is followed by step E33 for calculating a transformation, for each of the sub-blocks determined in step E41. Each of these sub-blocks is coded in the same way as a block An, m. The following steps are followed successively for each sub-block A1nm to AMln, m, which leads in particular to the determination of a transformation for the current sub-block, and to its storage if the criterion of step E34 is satisfied, or if the minimum sub-block size is reached (step E40).

 For each sub-block, if the criterion of step E34 is not satisfied, and if the minimum size of sub-block is not reached (step E40), the sub-block considered is itself divided to step E41 in sub-blocks of smaller size.

Thus, for example, in FIG. 11, the image fn is divided into twelve blocks An 1 to An 12 the blocks An, 3 and An, 6 are divided into four sub-blocks, and one of these sub-blocks A4n, 6 is divided into four sub-blocks A4 ln 6, A4,2,6, A4 3n 6 and
A4.4n, 6.

 As already specified, the coding of an image fn of the sequence results in a compressed image hn comprising a set of transformations (T ,,) associated with a quaternary tree.

The decoding algorithm corresponding to this embodiment is represented in FIG. 13 and comprises steps E200 to E300 respectively analogous to steps E20 to E30 of the algorithm described with reference to FIG. 9 with the following modifications:
the steps E220 and E270 of selecting the first block to be reconstructed ARn, m and of the following block, respectively, comprise the reading of the quaternary tree associated with the image, in particular to determine the size and the position of the block to be reconstructed.

 - Step E270 is followed by step E230 of dividing the reference image.

 These modifications are necessary to take account of the size and the position of the blocks during decoding; in particular, the size and the position of the source blocks depend on those of the block to be reconstructed.

 Another embodiment of the coding method according to the invention is shown in FIG. 14, and includes steps E51 to E67 stored in memory 23 (FIG. 3). Compared to the first embodiment, this embodiment consists in firstly finding, for each block to be coded, whether it can be coded by matching with a block of the reference image. If this coding is possible, the block is coded by block matching (in English: block matching); otherwise, the block is coded by means of a determined transformation as in the first embodiment.

 Steps E51 to E56 are identical to steps E1 to E6 previously described with reference to the first embodiment (FIG. 5).

 Step E56 is followed by step E57 to which the current block to be coded Anm is coded by block matching: a reference block Dni, mj which is most similar is sought in the reference image g1 possible for the block to be coded, according to a predetermined criterion, for example a measurement of distance d (Anm, Dn1mj) between the block to be coded and the correspondence block. As explained above, different types of distance can be implemented. The match block for which the criterion is minimized is selected.

 The next step E58 is a test to check whether the criterion is below a predetermined threshold. If the answer is positive, the relative position of the correspondence block Dni, mji in the reference image go 1, with respect to the position of the block to be coded An, m, in the image being coded fn, c ' that is to say the motion vector Vn mj, is stored in memory 25, in step E59. An indicator In m is also stored, to indicate that the block in question is coded by block matching. This indicator is for example an additional bit set to the value one. The motion vector Vnmj and the indicator 1n, m represent the coding information of the block An, m.

 Step E59 is followed by steps E63 to E67, respectively identical to steps E70 to E14 previously described.

If the answer is negative in step E58, that is to say if the coding by block matching does not give a satisfactory result, the step
E58 is followed by steps E60 to E67, respectively analogous to steps E7 to E14 previously described. Step E61 results in a transformation Tnm. The indicator In, m, for example the additional bit set to the value zero, is associated with each transformation Tn m.

 Alternatively, the two previous embodiments are combined. we first try to code the blocks by matching blocks, and then, if this first coding is not possible, by means of a transformation; the steps of coding by determining a transformation then include the variation of the size of the block to be coded according to a criterion.

 The decoding method corresponding to the coding in FIG. 14 is shown in FIG. 15 and includes steps E70 to E91 stored in memory 53 of the device 5 (FIG. 4).

Steps E70 to E73 are respectively identical to steps
E20 to E23 described with reference to Figure 9.

 The next step E74 is a test to determine whether the block under construction ARn, m has been coded by block matching. The test consists in analyzing the value of the indicator Inm and in searching if the indicator, m is at the value one.

 If the answer is positive, then the following step E75 consists in reading in the reference image gn-1 the correspondence block Dn 1 mj which is determined by the motion vector forming the compressed form Vn mj and in copying this block as an Arum decoded block. Step E75 is followed by steps E78 to E91 which are respectively identical to steps E25 to E30 previously described (FIG. 9).

 If the response to step E74 is negative, step E74 is followed by steps E76 to E91 which are respectively identical to steps E24 to E30 previously described (FIG. 9).

 Another embodiment of the coding method according to the invention is shown in FIG. 16 and includes steps E401 to E414 stored in memory 23 (FIG. 3).

 Compared to the first embodiment, this embodiment consists in dividing each image to be coded into zones, in analyzing the zones to determine whether they are fixed or mobile, then in coding the fixed zones according to a method of coding a fixed image. , and the mobile zones as in the first embodiment.

 Step E401 is the division of each of the images into predetermined zones Zon, 1 to Zn, El WHERE E is an integer equal to five by way of example. FIG. 17 represents the image fn divided into five zones Z, 1 to Zn, E which are bands of the same size.

 The next step E402 is an analysis of the zones to determine whether each of the zones Zne is "fixed" or "mobile". By fixed zone is meant an identical, or very little different, zone from the zone having the same location in the previous image in the sequence. Conversely, the area is said to be mobile if it is sufficiently different from the area having the same location in the previous image in the sequence. The difference between the area being processed and the area having the same location in the previous image is evaluated by measuring the distance between the two areas and comparing the distance to a threshold. As explained above, different types of distance are possible. A zone indicator lN is not associated with each of the zones to indicate whether the zone considered is fixed or mobile. The zone indicator lN n, e comprises for example a bit set to one or to zero to signify fixed zone or mobile zone.

 Step E402 is followed by step E403, which is an initialization at which the parameter n is initialized to one, to first consider the first image f1 of the sequence.

Step E403 is followed by step E404 in which the first image f1 of the sequence is coded according to a method of coding a fixed image. This method is for example a JPEG type method (Joint Picture Expert
Group), or the coding method described in the intilulated patent application "Method and device for coding and decoding data, in particular digital images" and filed on the same day as the present application by the applicant. The type of coding, in this case still image coding, is memorized by an indicator. This indicator is only memorized during coding and is not associated with the coded form of the areas of the image f1. The parameter n is incremented by one to then consider the following image f2.

 The next step E405 is an initialization of a parameter e to 1, to consider the first zone Zn 1 in the current image fn.

 The next step E406 is a test to determine whether the current zone Zn, e is fixed or mobile. The test consists of reading and analyzing the value of the zone indicator INn, e.

 If the zone being processed Zn, e is mobile, step E406 is followed by step E407 in which the zone Zn.e is divided into blocks, each block is then coded by a transformation Tnm between the block and source blocks taken from the reference image. Step E407 is equivalent to steps E2 and E5 to El 1 previously described (Figure 5).

 If the zone being processed Zn, e is fixed, step E406 is followed by step E408 which is a test to determine if the zone has already been coded during the coding of a previous image. This test consists in reading the coding type indicator of the zone considered in the previous image.

If the response to step E408 is negative, this step is followed by step E409 in which the zone Zne is coded according to the method of coding a fixed image, for example the method of the JPEG type (Joint Picture Expert
Group), or the coding method described in the intilulated patent application "Method and device for coding and decoding data, in particular digital images" and filed on the same day as the present application by the applicant. The coding type indicator indicates that the zone in question is coded according to a still image coding.

 If the answer is positive, the zone considered is not coded. Thus, a fixed area in several successive images is coded only once for the first of these images, then is not coded for the following images. As a result, coding is fast and requires few bits.

The next step E410 is a test to check whether e is equal to E, that is to say if the zone which has just been coded is the last zone to be coded of the current image fn. If at least one area remains to be coded in the fnl image, the step
E410 is followed by step E411 in which the parameter e is incremented by one, to consider the next area in the image fn. Step E41 1 is followed by step E406 previously described.

 When the zone which has just been coded is the last of the current image, step E410 is followed by step E412 which is a test to check whether the image fn which has just been compressed is the last of the sequence. If this is not the case, step E412 is followed by step E413 which is the construction of the mobile areas of the decompressed image gn A mobile area is reconstructed by the application of each of the transformations Tn m previously determined to the source blocks to form blocks In the name The mobile areas of the image gn are stored in memory 25 to be used for coding the mobile areas of the following image fun + 1.

Step E413 is followed by step E414 to increment the parameter n by one. Step E414 is followed by step E402 previously described. In the case where n = N, that is to say for the last image of the sequence, it is not necessary here to reconstruct the mobile areas of the image gN
When the response to step E11 is positive, all the images f1 to fun of the sequence have been compressed into compressed images h1 to hN.

 As a variant, the analysis of the zones is carried out as the images are coded.

 The decoding method corresponding to the coding in FIG. 16 is shown in FIG. 18 and includes steps E520 to E530 stored in memory 53 of the device 5 (FIG. 4).

 Step E520 is an initialization at which the parameter n is initialized at 1, to process the first compressed image h1 of the sequence.

 In the next step E521, the first compressed image h1 of the sequence is decoded according to a method which depends on the coding method of step E409 previously described. The parameter n is incremented by one to go to the next image h2.

 The next step E522 is the initialization of the parameter e to 1, to consider the first area to be reconstructed. The zones are preferably treated in the same order as with compression, which avoids having to memorize and process an index, or coordinates, to locate the zones.

 The next step E523 is a test to determine whether the current area has been coded as a fixed area or as a mobile area. The test consists in reading the value of the zone indicator IN, ..

If the current area is a fixed area, Step E523 is followed by step E524 in which it is tested whether the area having the same location in the previous image was decoded during the previous implementation of the step
E526.

If the area with the same location in the previous image has been decoded, this means that the current area is identical to an area already decoded. Step E524 is followed by step E525 in which this previously decoded area is copied as a decoded area for the current area of the current image. If the area having the same location in the previous image has not been decoded, step E524 is followed by step
E526 to which the current area is decoded according to a method which depends on the coding method. A decoded area indicator is memorized to be tested during a subsequent execution of step E524. Thus, an area which is fixed in several successive images is decoded for the first of these images then is copied for the following images.

If the current zone is a mobile zone, step E523 is followed by step E527 in which the mobile zone is decoded. Source blocks Dni, nn, i to Dn-l, m, K are selected in the reference image go 1, relative to each block to be reconstructed Art m of the current area. The reference image is selected in the same way as for coding, the transformation Tn, m defined by the functions an, m, k and bn, m is applied to the source blocks Dn1m1 to Dn1.m, K selected, according to the formula:
ARn, m = Zk (an, m, k X Dn-l, m, k) + bn, m.

 Step E527 results in the decoded blocks ARn, m of the current area, which are stored in memory 55.

 The next step E528 is a test to determine whether the zone which has just been decompressed is the last of the image being decompressed. If not, step E528 is followed by step E529 where

 Of course, the present invention is not limited to the embodiments described and shown, but encompasses, quite the contrary, any variant within the reach of ordinary skill in the art.

Claims (38)

 1. Method for coding a set of data representative of physical quantities, said set being composed of subsets (fn),  characterized in that it comprises, for each of the subsets, the steps of:  - division (E2) of the subset into data blocks to be coded (An, m) 7  - division (E6) of at least one reference subset (gn-1) into source blocks,  then, for each of the coded data blocks, the steps of:  selection (E7) of a predetermined number of source blocks (Dn 1, m, 1, i 1, m, K) in said at least one reference subset,  - determination (E8) of a transformation (Tnm) between the selected source blocks and the block to be coded.  2. Method for coding a set of data representative of physical quantities, said set being composed of subsets (fn),  characterized in that it comprises, for each of the subsets, the steps of:  - division (E52) of the subset into data blocks to be coded,  - division (E56) of at least one reference subset (go 1) into source blocks,  then, for each of the data blocks to be coded, the steps of:  - matching (E57) of the block to be coded with one of the source blocks, and, if a matching criterion is not satisfied (E58),  - selection (E60) of a predetermined number of source blocks (Dn 1, m, 1, i 1, m, K) in said at least one reference subset,  - determination (E61) of a transformation between the selected source blocks and the block to be coded.  3. Coding method according to claim 2, characterized in that the matching criterion of the matching step comprises a measurement of a first distance between the block to be coded (An, m) and the source block (Dni , mj).  4. Method for coding a set of data representative of physical quantities, said set being composed of subsets (fn),  characterized in that it comprises, for each of the subsets, the steps of:  - division (E401) of the subset into zones (Zn, 1 to Zn, E),  - determination (E402) of the type of each zone, from a first type of zone and a second type of zone, according to a criterion of zone,  - division (E407) of each second type area into data blocks to be coded (An, m),  - division (E407) of at least one reference subset (gn-1) into source blocks,  then, for each of the data blocks to be coded, the steps of:  - selection (E407) of a predetermined number of source blocks (Dn 1, m, 1, i 1, m, K) in said at least one reference subset,  - determination (E407) of a transformation (Tn, m) between the selected source blocks and the block to be coded.  5. Coding method according to claim 4, characterized in that it comprises the calculation (E406) of a second distance between each zone and a corresponding zone in an adjacent subset, in that the zone criterion consists in compare this distance to a threshold, and in that, if this distance is less than the threshold, the zone considered is of the first type, and of the second type otherwise.  6. Coding method according to any one of claims 1 to 5, characterized in that the transformation minimizes a third distance (d (Anm, Tn, m)) between the block to be coded (An, m) and its calculated approximation by applying the transformation to the selected source blocks (Dn.i, m, is Dni, m, K)  7. Coding method according to any one of claims 1 to 6, characterized in that the transformation is a multilinear approximation.  8. Coding method according to any one of claims 1 to 6, characterized in that the transformation is a multilinear approximation combined with a geometric transformation.  9. Coding method according to any one of claims 1 to 6, characterized in that the transformation is a polynomial on the values of the coefficients of the source blocks.  10. Coding method according to any one of claims 6 to 9, characterized in that the calculation of the third distance comprises the calculation of a difference between the values of the data of the block to be coded and the values of its approximation calculated in applying the transformation to the source blocks.  11. Coding method according to any one of claims 6 to 11, characterized in that the third distance is the mean square error calculated between the values of the block to be coded (An, m) and the values of the block resulting from l application of the approximation to the source blocks (Dn.i, m, i to Dni, mK).  12. Coding method according to any one of claims 1 to 11, characterized in that, for the first subset to be coded, said at least one reference subset (g0) is a predetermined subset.  13. Coding method according to any one of claims 1 to 11, characterized in that said at least one reference subset is a subset of the set.  14. Coding method according to any one of claims 1 to 11, characterized in that said at least one reference subset is a subset (go 1) of the set, which has been previously coded and then decoded .  15. Coding method according to any one of claims 1 to 14, characterized in that at least two reference subsets are considered, at least one of which is located among the subsets preceding the subset in coding course and at least a second is located among the subsets following the subset being coded.  16. Coding method according to any one of claims 1 to 15, characterized in that the source blocks (Dn1m1, Dn1, mK) are selected from a part of the reference subset (gn-1) which depends on the part in which the block to be coded (An, m) is located in the subset being coded (fn).  17. Coding method according to any one of claims 1 to 15, characterized in that the source blocks (Dn1, m, 1, Dn1mK) have a size which is determined as a function of the size of the block to be coded (An, m).  18. Coding method according to any one of claims 1 to 17, characterized in that the source blocks (Dn 1, m, 1, i 1, m, K) have a size multiple of a factor F2 of that of block to be coded (An, m) and are subsampled by the factor F, where F is an integer.  19. Coding method according to any one of claims 1 to 17, characterized in that the source blocks (Dn 1, m, 1, i 1, m, K) have the same size as the block to be coded (An, m).  20. Coding method according to any one of claims 1 to 19, characterized in that the block to be coded (Anm) is of predetermined size.  21. Coding method according to any one of claims 1 to 19, characterized in that the block to be coded (Anm) is of variable size, according to a size criterion.  22. Coding method according to claim 21, characterized in that it comprises the calculation (E38) of a fourth distance between the block to be coded and the approximation calculated by applying the transformation to the source blocks, in that that the size criterion consists in comparing (E39) the fourth distance to a threshold, and in that if the fourth distance is greater than the threshold, and if a minimum block size has not been reached, the block in progress coding is divided (E47) into blocks of smaller size.  23. Method for decoding a set of data representative of physical quantities, the set being composed of subsets (hn) which each include coded representations of data blocks, each of the blocks being coded by means of a transformation (Tn , m) between source blocks selected in at least one reference subset and the block,  characterized in that it comprises, for each of the blocks of each of the subsets to be decoded, the step of applying (E25) the transformation to the source blocks to decode the block (ARn, m).  24. Method for decoding a set of data representative of physical quantities, the set being composed of subsets (hn) which each include coded representations of data blocks,  characterized in that it comprises, for each of the blocks of each of the subsets to be decoded, the steps of:  - search (E74) if the block is coded by matching of blocks,  and, in the event of a negative response, the block then being coded by means of a transformation (Tn, m) between source blocks selected in at least one reference subset and the block:  - application (E77) of the transformation to the source blocks to decode the block.  25. A decoding method according to claim 24, characterized in that, in the event of a positive response to the search step, the block is decoded by copying a correspondence block determined by the matching of blocks.  26. A decoding method according to any one of claims 23 to 25, characterized in that it comprises the reading (E220, E270) of the size and the position of the blocks to be decoded.  27. Method for decoding a set of data representative of physical quantities, the set being composed of subsets (hn) which each include coded representations of areas belonging to a first or a second type of area, the coded representations of the second type of area being produced by data blocks,  characterized in that it comprises, for each of the blocks of the second type areas of each of the subsets to be decoded, the step of applying (E525) the transformation to the source blocks to decode the block.  28. Coding device (2) for a set of data representative of physical quantities, said set being composed of subsets (fn) l  characterized in that it comprises (24):  means for dividing the subsets (fn) into data blocks to be coded (An, m),  means for dividing at least one reference subset (go 1) into source blocks, for each of the subsets,  means for selecting a predetermined number of source blocks (Dn 1, m, 1, i 1, m, K) in said at least one reference subset (gon 1), for each of the data blocks to code (An m) of each of the subsets (fn), and  - Means for determining a transformation (Tnm) between the selected source blocks and the block to be coded.  29. Coding device (2) of a set of data representative of physical quantities, said set being composed of subsets (fn),  characterized in that it comprises (24):  means for dividing the subsets into data blocks to be coded (An, m),  means for dividing at least one reference subset (gon 1) into source blocks, for each of the subsets,  means for matching, for each of the data blocks to be coded, the block to be coded with one of the source blocks, and, if a correspondence criterion is not satisfied,  means for selecting a predetermined number of source blocks (Dn 1 m 1. Dni, m, K) in said at least one reference subset, for each of the data blocks of each of the subsets, and  means for determining a transformation (Tn, m) between the selected source blocks and the block to be coded.  30. Coding device (2) of a set of data representative of physical quantities, said set being composed of subsets (fn),  characterized in that it comprises (24):  - means for dividing the subset into zones (Zn1 to Zn E).  means for determining a transformation (Tn, m) between the selected source blocks and the block to be coded.  means for selecting a predetermined number of source blocks (Dn-1, m, 1. Dni, m, K) in said at least one reference subset, for each of the data blocks of the second type areas subsets, and  means for dividing at least one reference subset (gon 1) into source blocks,  means for dividing each second type area into data blocks to be coded (An m),  means for determining the type of each zone, from a first type of zone and a second type of zone, according to a zone criterion,  31. Coding device according to any one of claims 28 to 30, characterized in that the means of division, selection and determination are incorporated in:  - a controller (24),  - a read only memory (23) comprising a program for coding each of the data blocks, and  - a random access memory (25) comprising registers adapted to record variables modified during the execution of said program.  32. Decoding device (5) for a set of data representative of physical quantities, the set being composed of subsets (hn) which each include coded representations of data blocks, each of the blocks being coded by means of a transformation between source blocks selected in a reference subset (gj) and the block,  characterized in that it comprises means (54) for applying the transformation to the source blocks, for each of the blocks of each of the subsets to be decoded.  33. Decoding device (5) of a set of data representative of physical quantities, the set being composed of subsets (hn) which each include coded representations of data blocks,  characterized in that it comprises (54):  - search means, for each of the blocks (An, m) of each of the subsets to be decoded, if the block is coded by matching of blocks,  and in the event of a negative response, the block then being coded by means of a transformation (Tn, m) between source blocks selected in a reference subset and the block:  - means of applying the transformation to the source blocks to reconstruct the block (ARn, m).  34. Decoding device according to claim 32 or 33, characterized in that it comprises means for reading the size and the position of the blocks to be decoded.  35. Device for decoding a set of data representative of physical quantities, the set being composed of subsets (hn) which each include coded representations of zones belonging to a first or a second type of zone, the coded representations of the second type of area being carried out by data blocks,  characterized in that it comprises means for applying the transformation to the source blocks, for each of the blocks of the second type areas of each of the subsets to be decoded.  36. Decoding device according to any one of claims 32 to 34, characterized in that the application means are incorporated in:  - a controller (54),  - a read only memory (53) comprising a program for decoding each of the data blocks, and  - A random access memory (55) comprising registers adapted to record variables modified during the execution of said program.  37. Digital apparatus comprising means suitable for implementing the coding method according to any one of claims 1 to 22, and / or means suitable for implementing the decoding method according to any one of claims 23 to 27.  38. Digital apparatus comprising the coding device according to any one of claims 28 to 31, and / or the decoding device according to any one of claims 32 to 36.