FR2862168A1 - Digital image sequence flow regulation method for digital camera, involves suppressing digital data among digital data constitutive of image based on comparison of maximum flow of compressed image with target flow - Google Patents

Abstract

The method involves determining a target flow for an image, and determining a quantization step corresponding to the flow. The image is compressed by quantizing using the quantization step, and maximum flow of the compressed image is compared with the target flow. Digital data among digital data constitutive of the image is suppressed based on the comparison result, to make the maximum flow equal to the target flow. Independent claims are also included for the following: (a) a digital image sequence flow regulation device (b) a communication apparatus comprising a digital image sequence flow regulation device (c) an information storing unit comprising instructions for execution of digital image sequence flow regulation method (d) a computer program loadable in a programmable apparatus comprises instructions for implementing digital image sequence flow regulation method.

Description

The present invention relates to a method and a device for

  flow control of a sequence of digital images.

  At present, digital cameras compress still images in JPEG format and also offer to compress Motion-JPEG format video (MJPEG).

  This format, not standardized, consists of compressing each image of the sequence in JPEG format.

  Thus, the same compression module is used for both still images and video sequences.

  JPEG2000 Part 1 (ISOIIEC 14495-1, JPEG2000 Image Coding Standard, Part 1, 2000) has recently been approved as a new international standard and has many features that are not available with the old JPEG standard.

  For example, in the case of a digital camera user, it is interesting to be able to convert an image from a given quality mode to a lower quality mode image, which means that the definitive quality mode of an image can be chosen after taking the photo. This feature is not available with the JPEG standard while it is available with the JPEG2000 standard.

  It is therefore interesting to use a JPEG2000 compression module in digital cameras.

  The JPEG2000 compression module can therefore be used to compress Motion-JPEG2000 (MJ2K) format video sequences, each image in the sequence being compressed in JPEG2000 format.

  In view of the foregoing, it would be useful to have a data rate control system for obtaining a compressed video stream at constant or near-constant rate.

  Obtaining such a bitrate would make it possible to respect the limitations of a digital camera such as the speed of writing data.

  There is a set of methods for controlling the bit rate of video sequences for various formats such as MPEG-2, MPEG-4, H263, H26L.

  The methods called CBR (Constant Bit Rate), which aim to obtain a quasi-constant bit rate, often use mathematical models to connect the distortion, the bit rate and the quantization step.

  Thus, for a given target bit rate, the quantization step to be used for the compression of an image can be estimated more or less precisely.

  A method is known that provides for the use of a simple model linking the quantization step and the bit rate of an image, and to adapt the parameters of the model according to the results obtained for the previously compressed images.

  This method applied to MPEG-4 video sequences is described in the article "A new rate control scheme using quadratic rate distortion model" by Chiang and Y.Q. Zhang in IEEE Transactions on Circuits and Systems for Video Technology, Vol. 7, No. 1, February 1997 and in US Pat. No. 6,243,497 entitled "Apparatus and method for optimizing the rate control in a coding system".

  However, a limitation of this method occurs during the scene changes that occur between the images of the video sequence: the parameters of the model that have been calculated from the previous images are no longer suitable, thus leading to variations in rates.

  Other methods try to predict the content changes in the video sequence to estimate compression complexity.

  Nevertheless, preliminary calculations on the raw images are then necessary, which requires for the implementation of these methods of high computational capabilities and / or increases the time required to complete the compression process.

  Such methods are described, for example, in the article "A new MPEG-2 rate contract scheme using scene change detection" by Park, Lee and Chang in ETRI Journal, in the article entitled "A Software-based High-Quality MPEG-2 Encoder Employing Scene Change Detection and Adaptive Quantization ", International Conference on Consumer Electronics (ICCE), June 2001 by D. Farin, N. Mache, PHN and US Pat. No. 5,872,598 entitled "Scene change detection using quantization scale factor rate control" and US Pat. No. 6,097,757 entitled "Real-time variable bit rate encoding of video sequence employing statistics".

  The methods proposed above do not provide a flow control system for quickly obtaining compression of a constant or near constant rate image sequence.

  The methods proposed above do not make it possible to obtain the compression of a constant or quasi-constant flow sequence without, however, causing a computationally high cost.

  The present invention aims to remedy at least one of the aforementioned drawbacks by proposing a method for regulating the bit rate of a sequence of digital images, characterized in that it comprises, for a current image of the sequence, the following steps: E2 of a target rate Rc (i) for the current image, - E4 determination of a quantization step Q; corresponding to the target rate Rc (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the flow rate, - compression E5 of the current image including in particular a quantization step of the image with the quantization step Q; previously determined, - comparison E10 of the maximum rate Rmax (i) obtained for the current image thus compressed with the target rate Rc (i) in order to determine the largest of the two rates, - depending on the result of the comparison, decision as to possible suppression of digital data among the digital data constituting the current image in order to bring the value of the rate Rmax (i) as close as possible to the target rate Rc (i).

  Correlatively, the invention also relates to a device for regulating the bit rate of a sequence of digital images, characterized in that it comprises for a current image of the sequence: means for determining a target bit rate Rc ( i) for the current image, - means for determining a quantization step Q; corresponding to the target rate Rc (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the flow rate, - means for compressing the current image including means for quantizing the image with the quantization step Q; previously determined, - means for comparing the maximum rate Rmax (i) obtained for the current image thus compressed with the target rate Rc (i) in order to determine the greater of the two rates, - decision means as to a possible deleting digital data from the digital data constituting the current image in order to bring the value of the rate Rmax (i) as close as possible to the target rate R, (i), as a function of the result of the comparison.

  Thus, the proposed rate regulation makes it possible, by comparing the maximum rate obtained with the target bit rate, to make a decision on the necessity of implementing a bit rate / distortion allocation in order to obtain a fast convergence towards the target bit rate.

  It is thus possible to obtain convergence towards the target bit rate with three or four images.

  When compressing the first images of the sequence, the model is based on a small number of images already compressed and therefore has little information to determine the parameters of the model.

  The bit rate-distortion allocation is then useful to respect the target bit rate of the current image when the quantization step used is too small. The bit rate-distortion allocation also makes it possible to obtain a better distribution of the bit rate on the following images: it is this better distribution which allows a faster convergence of the model.

  It should be noted that in the absence of a bit rate / distortion allocation, and in the case of a quantization step that is too small for the current picture, a lower target bit rate would be obtained for subsequent pictures of the sequence. which would not allow to respect the constraint of constant or quasi-constant flow.

  When the converged model then has an optimized quantization step, which makes it possible not to deal with superfluous data during compression and, in particular, during entropy coding.

  When the model converges, it is also decided to dispense with a rate-distortion allocation, which represents a saving of time.

  According to one characteristic, the method comprises a data deletion step E11 when the maximum rate Rmax (i) is greater than the target rate Rc (i).

  According to one characteristic, the method comprises a comparison step E8 of the proximity of the maximum rate Rmax (i) with the target rate Rc (i).

  According to another characteristic, the method comprises a decision step concerning a possible reinitialization of the quantization step prediction model for the next image of the sequence, as a function of the result of the proximity comparison of Rmax (i) and Re for the current image, the decision to reinitialize the model intervening when the model is no longer adapted to the current image.

  Resetting the model makes it possible to obtain constant or near-constant image compression even when scene changes have occurred between the images of the sequence.

  More particularly, the method includes a step of resetting the quantization step prediction model for the next image of the sequence when the relative difference between Rmax (I) and Rc (i) exceeds a predetermined threshold.

  According to one characteristic, the decision to reset the quantization step prediction model depends on the number of images that have been previously compressed since the last reset of the model.

  A minimal number of images is indeed necessary to reach the convergence of the model: after reinitialization of the model, one does not make a new reinitialization before compressing this minimal number of images.

  More particularly, the method includes a step of resetting the quantization step prediction model for the next image of the sequence when the relative difference between Rmax (i) and Rc (i) exceeds a predetermined threshold and the number of quantization steps. images previously compressed since the last reset of the model reaches a predetermined minimum number N. Thus, the proximity of the maximum bit rate with respect to the target bit rate is taken into account in order to determine whether the model is still adapted to the current picture or not.

  According to one characteristic, the method comprises a step of updating E3 of the parameters of the quantization step prediction model prior to the step of determining the quantization step of the current image.

  Thus, for each new image of the sequence, the prediction model is refined.

  More particularly, the step of updating the model parameters for the current image takes into account the pairs of values Rmax (i), Q; obtained during the compression of the images of the sequence preceding the current image.

  The model is thus refined taking into account in addition to the preceding images, for each new image, the image which precedes it in the sequence.

  More particularly, the step of updating the model parameters for the current image takes into account the nmod pairs of values Rmax (i), Q, obtained during the compression of the nmod images of the sequence preceding the current image. and that have been compressed since the last reinitialization of the model, the reinitialization of the model occurring when the model is considered no longer suitable for a current image.

  According to one characteristic, when the number nmod of images of the sequence which have been compressed since the last reset of the model has reached a predetermined maximum number N ,, then the step of updating the parameters of the model for each of the following current frames of the sequence take into account the N, N pairs of values Rmax (i), Q; obtained during the compression of N, images preceding the current image considered.

  Thus, beyond a certain number of processed images, it is considered that it is no longer necessary to continue to take into account the first images to update the model.

  Indeed, the first images are too old to provide a model adapted to more recent images in the sequence.

  According to one characteristic, when the relative difference between Rmax (i) and Rc (i) exceeds a predetermined threshold for the current image of the sequence, the step of updating the parameters of the model for the next image of the sequence corresponds to a model reset step that takes into account the m pairs of values Rmax (i), Q; obtained during the compression of the last m frames of the sequence, the number m being determined by the minimum number of pairs of values Rmax (i), Q; necessary to determine the parameters of the model.

  Thus, in case of reinitialization of the model, one is no longer interested in images that are too old to proceed to the determination of the parameters of the model.

  According to one characteristic, when the prediction model of the quantization step of the current image does not provide a mathematical solution to the determination of the quantization step, then the quantization step is determined by averaging the quantization steps Q; images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, a reinitialization of the model intervening when it is no longer adapted to the current image.

  According to one characteristic, the model for predicting the quantization step of the current image is approximated by the equation R = aQ "1+ bQ_2, R 2862168 8 and Q respectively denoting the bit rate and the quantization pitch for the current image. and a and b being the parameters of the model.

  More particularly, the image sequence complies with the Motion JPEG2000 standard and the data deletion involves a post-compression distortion rate allocation algorithm.

  According to another aspect, the invention also relates to a method for regulating the bit rate of a sequence of digital images, characterized in that it comprises, for a current image of the sequence, the following steps: determination E2 of a target bit rate R, (i) for the current image, determination E4 of a quantization step Qi corresponding to the target rate Rc (i) for the current image, the determination of the quantization step using a mathematical model for predicting the pitch of quantization as a function of the flow rate; compression E5 of the current image notably comprising a step of quantizing the image with the quantization step Q; previously determined, - comparison of the proximity of the maximum rate Rmax (i) obtained for the current image thus compressed with the target rate Rc (i), - depending on the result of the comparison, decision on a possible reset of the predicting the quantization step for the next image of the sequence, the decision to reinitialize the model occurring when the model is considered no longer suitable for the current image.

  Correlatively, the invention also relates to a device for regulating the bit rate of a sequence of digital images, characterized in that it comprises for a current image of the sequence: means for determining E2 of a target bit rate Rc ( i) for the current image, - determination means E4 of a quantization step Q; corresponding to the target rate Rc (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the flow rate, - compressing means E5 of the current image including notably means for quantizing the image with the quantization step Qi previously determined; means for comparing the proximity of the maximum bit rate Rmax (i) obtained for the current image thus compressed with the target bit rate Rc (i); decision means concerning a possible reinitialization of the quantization step prediction model for the next image of the sequence, according to the result of the comparison, the decision to reset the model occurring when the model is considered to be no longer suitable to the current image.

  The invention thus makes it possible, by comparing the maximum bit rate obtained with the target bit rate, to check the validity of the model parameters with a very reasonable calculation time.

  Indeed, we decide whether or not to adapt the model for a current image of the sequence, which makes it possible, in case of adaptation of the model, not to take into account for the estimation of the model previously compressed images not presenting no similarity with the current image. Thus, an effective estimate of the quantization step is obtained. During the compression step, only the necessary data are then taken into account.

  The invention thus avoids resorting, as in the prior art, to methods with high computational costs which aim at predicting changes in content in the sequence.

  The invention also relates to a communication apparatus comprising a device as briefly described above.

  According to another aspect, the invention also aims at: a computer-readable information storage means or a microprocessor comprising code instructions of a computer program for executing the steps of the method according to the invention as briefly set forth above, and - a partially or fully removable computer-readable information storage medium or microprocessor having code instructions of a computer program for executing the steps of method according to the invention as briefly discussed above.

  According to yet another aspect, the invention relates to a computer program loadable in a programmable apparatus, comprising sequences of instructions or portions of software code to implement steps of the method according to the invention as briefly discussed below. above, when said computer program is loaded and executed on the programmable apparatus.

  The characteristics and advantages relating to the device, to the apparatus comprising such a device, to the information storage means and to the computer program being the same as those described above concerning the method according to the invention, it will not be possible to not remembered here.

  Other characteristics and advantages of the present invention will emerge more clearly on reading the description which follows, made with reference to the appended drawings, in which: FIG. 1 schematically represents a programmable apparatus implementing the invention; ; FIG. 2 schematically represents a compression system of a digital image signal in which the invention is implemented; FIG. 3a schematically represents a digital image at the output of the image source 210 of FIG. 2; FIG. 3b schematically represents the image of FIG. 3a after having been transformed by the circuit 221 of FIG. 2; FIG. 4 schematically illustrates the representation of the bit planes of the coefficients of a data block B; ; FIG. 5 is an algorithm for regulating the bit rate of an image sequence according to the invention; FIG. 6 is an algorithm forming part of the algorithm of FIG. 5 and detailing step E1 of the latter; FIG. 7 illustrates the evolution of the bit rate of a known compressed video sequence as a function of the images of the sequence.

  According to the embodiment shown in FIG. 1, an apparatus embodying the invention is, for example, a digital photographic apparatus 100 connected to various peripherals such as a means of acquiring or storing images. sequence of images.

  The apparatus 100 of FIG. 1 comprises a communication bus 108 to which are connected: a central unit 120 (microprocessor), a read-only memory 122, including a program "Progr" enabling the programmable apparatus to implement the invention (although only one program is identified, it is possible to have several programs or sub-programs to implement the invention), - a random access memory 124, including registers adapted to record modified variables during the execution of the aforesaid program, - a flash memory 126 (card) for storing compressed images, - an image capture circuit 128, for converting an optical image into a digital signal, - a color processing circuit 130 which converts the digital signal and transmits it to the image memory 132, - a liquid crystal display 134 for displaying an image stored in the flash memory 126 or in the image memory 132, - a control circuit display element 136, for controlling the display of the liquid crystal display 134 under the control of the central unit 120, - switching means 138 (open / close switch, conversion mode switch, switch power, image selection switch, etc.) used by the user of the digital camera, - an input port 140 for receiving an input signal provided by the switching means, - a cable interface 142 used to connect a host interface 144 of the apparatus to a corresponding digital camera interface in host computer 146 (cable 142 is, for example, in accordance with the USB specification, meaning "Universal Serial Bus" "in Anglo-Saxon terminology, and it serves to repatriate the images contained in memory).

  The communication bus 108 allows communication between the various elements included in the apparatus 100 or connected thereto. The representation of the bus is not limiting and, in particular, the central unit is able to communicate instructions to any element of the apparatus 100 directly or via another element of the apparatus 100.

  According to one variant, the memory 126 may contain compressed and stored data as well as the code of the invention (program "Progr") which, once read by the apparatus 100, will be stored in the memory 124.

  According to another variant, the program can be received via a communication network to be stored identically to that described above.

  In general terms, an information storage means, readable by a computer or by a microprocessor, integrated or not integrated into the apparatus and possibly removable, stores a program whose execution implements the method according to the invention.

  More generally, the program can be loaded into one of the storage means of the apparatus 100 before being executed.

  The central unit 120 will execute the instructions relating to the implementation of the invention, instructions stored in the read-only memory 122 or in the other storage elements. When powering up, the program or programs, which are stored in a non-volatile memory, for example the read-only memory 122, are transferred into the random access memory 124 which will then contain the executable code of the invention, as well as registers to memorize the variables necessary for the implementation of the invention.

  It should be noted that the communication apparatus which is able to implement the invention can also be a programmed apparatus.

  According to an embodiment chosen and shown in FIG. 2, a data coding or compression system is a system which comprises an input 200 to which a source 210 of uncoded data is connected.

  The source 210 comprises, for example, a memory means, such as random access memory, hard disk, diskette, compact disk, for storing uncoded data, this memory means being associated with an appropriate reading means for reading the data therein. A means for storing the data in the memory means may also be provided.

  It will be considered more particularly in the following that the data to be encoded are a series of digital samples of origin representative of physical quantities and representing, for example, an IM image.

  The source 210 supplies a digital image signal IM at the input of the compression circuit 220. The image signal IM is a series of digital words, for example bytes. Each byte value represents a pixel of the IM image, here at 256 gray levels, or black and white image. The image may be a multispectral image, for example, a color image having red-green-blue or luminance-chrominance components. Either the color image is processed in its entirety, or each component is processed similarly to the monospectral image.

  It will also be appreciated that the image may be a still image (eg, photograph) or an image of a sequence of images.

  Coded or compressed data exploitation means 300 are connected at the output 225 of the compression system 220.

  The user means 300 comprise, for example, coded data storage means and / or coded data transmission means.

  The compression system 220 conventionally comprises, from the input 200, a spatio-frequency transformation circuit 221 which can implement one or more spatio-frequency transformations. For example, the circuit 221 implements decompositions into frequency subband signals of the data signal, so as to perform an analysis of this signal.

  The transformation circuit 221 is connected to a quantization circuit 222. The quantization circuit implements a quantization known per se, for example a scalar quantization, or a vector quantization, coefficients, or groups of coefficients, signaling signals. frequency subbands provided by circuit 221.

  The circuit 222 is connected to an entropy coding circuit 223, which performs an entropy coding, for example a Huffman coding, or an arithmetic coding, of the data quantized by the circuit 222.

  The circuit 223 is connected either to a device 224 which performs a bit rate / distortion allocation on the data thus compressed, or directly to the output 225 (as indicated by the dashed reference 226) when no bit rate / distortion allocation is available. decided.

  We will return to the rate / distortion allocation in the description of FIG.

  For an image sequence conforming to the Motion-JPEG2000 standard ("MJ2K"), each image in the sequence is compressed according to the JPEG2000 standard.

  The steps performed in the circuits 221, 222 and 223 of FIG. 2 illustrate the steps of a still image encoder such as an encoder according to the JPEG2000 standard. FIG. 3a schematically represents a digital image IM in

  output of the image source 210 of FIG.

  This image is decomposed by the transformation circuit 221 of FIG. 2 which is a dyadic decomposition circuit with three decomposition levels.

  The circuit 221 is, in this embodiment, a conventional set of filters, respectively associated with decimators by two, which filter the image signal in two directions, into sub-band signals of high and low spatial frequencies. The relationship between a highpass filter and a lowpass filter is often determined by the conditions of perfect signal reconstruction. It should be noted that the vertical and horizontal decomposition filters are not necessarily identical, although in practice this is generally the case. The circuit 221 here comprises three successive analysis blocks for decomposing the IM image into subband signals according to three levels of decomposition.

  In general, the resolution of a signal is the number of samples per unit of length used to represent that signal. In the case of an image signal, the resolution of a subband signal is related to the number of samples per unit length used to represent that subband signal horizontally and vertically. The resolution depends on the number of decompositions made, the decimation factor and the resolution of the initial image.

  The first analysis block receives the digital image signal IM and, in known manner, outputs four sub-band signals LL3, LH3, HL3 and HH3 of the highest resolution R3 in the decomposition.

  The subband signal LL3 comprises the components, or samples, of low frequency, in both directions, of the image signal. The subband signal LH3 comprises the low frequency components in a first direction and high frequency in a second direction of the image signal. The subband signal HL3 comprises the high frequency components in the first direction and the low frequency components in the second direction. Finally, the subband signal HH3 comprises the high frequency components in both directions.

  Each subband signal is a set of real samples (it could also be integers) constructed from the original image, which contains information corresponding to an orientation respectively vertical, horizontal and diagonal of the content of the image, in a given frequency band.

  The subband signal LL3 is analyzed by an analysis block similar to the previous one to provide four LL2, LH2, HL2 and HH2 sub-band signals of resolution level R2.

  Each of the resolution subband signals R2 also corresponds to an orientation in the image.

  The subband signal LL2 is analyzed by an analysis block similar to the previous one to provide four subband signals (by convention), LH1, HL1, and HH1 of resolution level Ri. It will be noted that the sub-band LLo alone forms the resolution Ro.

  Each of the RI resolution subband signals also corresponds to an orientation in the image.

  FIG. 3b represents the IMD image resulting from the spatial-frequency transformation applied to the IM image, by the circuit 221, in ten sub-bands and in four resolution levels: Ro (LLo), R1 (LL2), R2 (LL3), R3 (original image). The IMD image has as much information as the original image IM, but the information is frequently distributed according to three levels of decomposition.

  Of course, the number of decomposition levels, and therefore of subbands, can be chosen differently, for example 16 subbands in six resolution levels, for a two-dimensional signal such as an image. The number of subbands per resolution level may also be different. In addition, the decomposition may not be dyadic. The analysis and synthesis circuits are adapted to the size of the processed signal.

  In FIG. 3b, the samples resulting from the transformation are ranked subband by subband.

  It will be noted that each sub-band of the IMD image is partitioned into blocks Bi, some of which are represented in FIG. 3b.

  The circuits 222 and 223 of FIG. 2 apply independently to each block of each subband considered, the device 224 applying to all the blocks. The image signal compressed by the system 220 thus conveys blocks of samples obtained by coding the original samples and which constitute a data stream (known in English terminology as the term "bitstream").

  When the image signal complies with the JPEG2000 standard, these sample blocks are known in English terminology as "codeblocks".

  The coded picture signal also has a header.

  This header includes information about the size of the image, namely its width and height, its position in a reference frame, and the number of resolutions.

  The size of the blocks for each subband at a given resolution is determined by two parameters indicated by markers in the bit stream of the image signal conforming to the JPEG2000 standard.

  Note that it is possible not to include in the signal blocks corresponding to at least one level of resolution such as, for example, the maximum resolution level R3 or even several higher levels.

  The suppression of a resolution during the rate / distortion allocation of FIG. 2 is accompanied by a reduction by two of the size of the image, which is advantageous in terms of the memory space freed to be able, by for example, memorize additional signals.

  Such an advantage is particularly advantageous for a device of limited memory capacity and which stores digital images, such as for example a digital camera.

  It should be noted that by keeping in the signal to be exploited only the blocks of the lower resolution level Ro (low sub-band LLo), a good representation on a reduced scale of the original image signal will nevertheless be obtained.

  FIG. 4 illustrates the bit plane entropic coding of a data block B; an image signal as performed by circuit 223 of FIG. 2. Such progressive coding is described in the article entitled "High scalable performance image compression with EBCOT" by D. Taubman published in "IEEE Transactions on image processing ", vol. 9, No. 7, July 2000, pages 1158 to 1170.

  Each coefficient of a block B; is a real number which is quantized on several bits (sequence of bits), for example five bits in Figure 4.

  The bit plane PB1 contains the most significant bits, called MSB (known in English terminology as the "Most Significant Bit"), of the 2862168 18 coefficients of the block B. The bit planes PB2 to PB5 respectively contain the bits. less significant coefficients of the block B. The bit plane PB5 thus contains the least significant bits, called LSB (known in English terminology under the term "Less Significant Bit"), the coefficients of the block B. Each bit plane is coded in several coding passes, for example two. The result of the first coding pass provides a portion of the coding data for the block in question, and the result of the second pass provides another portion of the coding data with additional details.

  In order to regulate the bit rate for an image, it is possible to include in the final bitstream only a part of the data (bit portions) corresponding to an integer number of coding passes: each coding pass thus corresponds to at a possible truncation point of the bit stream for the block considered. Each coding pass is associated with a bit / distortion torque which corresponds to the additional bit rate and the overall distortion decrease for the reconstructed picture when the corresponding data just coded according to the coding pass considered are included in the bit stream. final.

  The rate / distortion allocation corresponds to the selection, for each of the blocks of the image, of one or more points of truncation of the data T; (R, ", distortion D ') of the bit stream of block B, as shown in FIG. 4.

  For example, one may choose to use the PCRD post-compression rate / distortion allocation algorithm described in the JPEG2000 standard or in the book "JPEG2000 Image compression fundamentals, standards and practice" by David S. Taubman and Michael W Marcellin.

  The PCRD (Post Compression Rate-Distortion Optimization) algorithm makes it possible to optimize the truncation points (R, ", D, ''), for example, to minimize the total distortion D => D;" of the image with a flow constraint R = ER, n 'Rx, or else to minimize the total flow rate R = ER, n' of the image with a distortion stress D = EDF 'D.

  This algorithm with a constraint on the bit rate is, for example, used in the present description for the bit-distortion allocation.

  FIG. 5 illustrates a flow control algorithm of a digital image sequence, such as an MJ2K sequence, according to the invention.

  The algorithm comprises several steps or portions of software code corresponding to steps of the method according to the invention and whose execution, for example, by the central unit 120 of the apparatus of FIG. 1, allows the implementation of the process.

  The computer program "Progr" which, as previously mentioned, is stored, for example, in the memory 122 (FIG. 1), is based on this algorithm and the execution of this program makes it possible to implement the method according to FIG. 'invention.

  The algorithm comprises a first initialization step E1 which is more particularly detailed in FIG.

  Step El will make it possible to estimate the quantization step for the first two images of the sequence, as will be seen later.

  During step E2, a target rate Rc (i) for the current image i is determined by calculation.

  To do this, we consider a group of images called GOP (meaning "Group Of Pictures" in English terminology), for which the debit will be respected to the extent that we can not guarantee that the target rate will be reached for each picture.

  If the target rate for the GOP is Rrota, the target rate R, for the image number n of the GOP image group, is calculated as follows: n IR, o, al E R1 R _ i = l NcoP + ln 'with NcoP which designates the total number of images in a GOP, and Ri designating the rate reached for the compressed image j.

  The following step E3 provides an update of the parameters of a mathematical model for predicting the quantization step as a function of the flow rate. During this step, the parameters of the model used to predict the quantization step corresponding to a flow constraint Rc (i) are determined.

  The model used to determine the quantization step is the Validation Model 5 (VM5) of the MPEG-4 standard.

  In the exemplary embodiment described here, this model is, for example, adapted to image sequences compliant with the JPEG2000 standard.

  It is assumed that the data source, ie the coefficients resulting from the wavelet transformation, has a Laplacian distribution, as indicated below: P (x) = 2 e alx If a measure is used distortion value defined by D (x, x) _ x -, a simple solution for the rate / distortion function is given by the following formula: R (D) = 14 J aD This function is developed in Taylor series as follows: 1 \ 1 1 \ 2 R (D) = \ -1-2 -1 + R3 The distortion measure is assumed to be proportional to the quantization step for an image, and the function is approximated to the second order. We thus obtain the mathematical model of prediction of the following quantization step: R = aQ- '+ bQ-2 Cl) where Q denotes the quantization step and a and b are the parameters of the model. This approximated model is valid for a still image.

  Nevertheless, if we assume that the successive images of a video sequence are similar, we can extend this model to a sequence or at least a portion of a sequence according to the similarity of the images of the sequence.

  The parameters a and b are then determined from the pairs (R, Q) obtained during the compression of the preceding images. These parameters are thus updated in step E3 for each new image by adding the pair (R, Q) corresponding to the last compressed image: nt odIR. ## EQU1 ## where nmod is the number of previously compressed images which is used to estimate the parameters of the model, and Qj, Rj respectively denote the quantization step and the associated bit rate for the compressed image j, j varying from 0 to nmod-1. More particularly, nmod is the number of images compressed since the last reinitialization of the model. The reinitialization of the model occurs, as described later, when the model is considered no longer suitable for a current image.

  In the next step E4, the quantization step for the current image is determined by solving equation (1) above: 2b Q = if a2 + 4bRe> 0, Vat + 4bR, a (4) otherwise. (5) nmod

  In the case where a2 + 4bR, <0, the model does not provide a possible mathematical solution. It is therefore chosen to take as quantization step value the average of quantization steps Q; images of the sequence preceding the current image and which have been compressed (equation (5) above).

  In order to avoid abrupt quality variations between two successive images of a sequence, the value of the quantization step can be limited according to the quantization step of the previous image in the sequence, as follows: 1 s <Q <1 + s, where s is the percentage of Qr- difference allowed between two quantization step values for two successive images.

  We have then: Q, = max (1 s) xmin Q ;, (1+ s) x The quantization step thus adapted to the current image is applied to the frequency subband LL, whereas the value of the steps Quantification for other sub-bands is deduced using, for example, the recommendations of the standard to take into account the human visual system (ISO / IEC 14495-1, JPEG2000 Image Coding Standard, Part 1, 2000, Appendix J).

  During the next step E5, the current image is compressed according to the JPEG2000 standard, in the example considered. This step comprises the following steps applied to the current image: wavelet transform, quantization with the quantization step obtained in step E4 and entropy coding.

  The next step E6 is a test on the value of the number n, nod of images already compressed since the last reinitialization of the model and which are used for the determination of the model.

  If i2n, od is less than a predetermined maximum value NH, then the next step E7 increments the value of nn, od by one, otherwise it goes directly to step E8.

  The two steps E6 and E7 make it possible to have a "sliding window" of NW images of width to determine the parameters of the model.

  Thus Nry (for example equal to 30) is limited to the number of previously compressed images contributing to the model (corresponding NW couples R, Q).

  Indeed, even without a change of scene, images that are very far in time may have significant differences and will therefore no longer be useful for determining the model.

  In the next step E8, the proximity of the maximum flowrate reached R is compared. (i) obtained for the compressed current image i with the target rate Mi).

  More particularly, the absolute value of the relative difference between R 1, (i) and Re (i) is calculated and this value is compared with a predetermined threshold value.

  If the calculated value is less than the threshold value, this means that the model has made it possible to satisfactorily estimate the quantization step to reach the target bit rate and it is therefore decided to go directly to the next step E10.

  If, on the other hand, the calculated value is greater than the threshold value, then the model does not make it possible to satisfactorily estimate the quantization step and is therefore no longer adapted to the current image.

  It is further necessary that the number of compressed images used to decide on the reinitialization of the model be greater than or equal to a predetermined value N (for example set at 4) to allow the convergence of the model.

  In this case, it is therefore decided to go to the next step E9.

  After the decision step E8, the step E9 is a step of resetting the model: in fact, the relative difference obtained between the target bit rate and the achieved bit rate for the current picture means that the parameters of the model, determined from previous images, are no longer suitable. It is therefore supposed that a scene change has occurred in the sequence and the model is reinitialized by considering only the last two compressed images to determine the parameters a and b, that is to say that we set nn , od has 2.

  Given the model considered in this exemplary embodiment, two images and therefore two pairs of values (Rmax (i), Qi) obtained during the compression of two images are sufficient to estimate the quantization step of an image with the model. . If one adopts another model, the number m of images to be taken into account to use the model corresponds to the minimum number of couples (Rmax (i), Q;) necessary to determine the parameters of the model.

  The reinitialization of the model occurs during the processing of the next image (i + 1) of the sequence, in step E3 of updating the model.

  The parameters a and b are then determined using equations (3) and (2) respectively, using the last two compressed images of the sequence.

  In the next step El0, a test is performed to determine whether the maximum rate Rmax (i) for the current image is greater than the target rate Re (i) determined.

  If not, it is decided to go to step E12 and thus not to delete any of the data constituting the current image.

  If so, it is decided to delete data from the current image in order to bring the value of the maximum flow rate reached R, (i) as close as possible to the target flow rate Re (i) and thus to Step El 1.

  Step E11 is a rate / distortion allocation step which, in the example considered, is specific to the JPEG2000 standard.

  As described with reference to FIG. 4, data is suppressed to reach the target rate Mi) using the PCRD algorithm.

  In the next step E12, a test is made to determine if the current image is the last frame of the sequence.

  If so, the process of regulating the bit rate of the image sequence is terminated.

  Otherwise, proceed to step E13 which plans to increment by one unit the image number i in order to move to the next image of the sequence and it is looped on step E2 previously described where the we will determine the target rate of the new image.

  As mentioned above, the initialization step E1 refers to the

Figure 6 for its description.

  Figure 6 is an algorithm that illustrates the quantization step estimation for the compression of the first two frames of the sequence. Indeed, an initialization step is necessary since at least two pairs (Q, R) are necessary to estimate the parameters a and b of the model with the equations (2) and (3) referred to above.

  Three quantization step values Q ,,, ,, ,,, Q ,,, ed, Q, ax are thus fixed as follows: <Q ,,, ed <Q. and which are such that a compressed image with the quantization step Qn, ax will have a lower bit rate and quality than this same image encoded with the paso ,,, ,,,. The three normalized quantization step values for the LL subband during JPEG2000 compliant compression are, for example, 0, 0126, 0.0161, and 0.022.

  The algorithm comprises a first step E101 in which the value of i (image number) is initialized to 0 and the quantization step Q is fixed; to the Qmed value.

  In the next step E102, the target rate Rc (i) for the current image is determined as described in step E2 of the algorithm of FIG. 5.

  The next step E103 corresponds to the compression of the current image according to the JPEG2000 standard, in the example considered. This step comprises the following steps: wavelet transform, quantization with the quantization step obtained in step E101 (or steps E106, E109 when i 0) and entropy coding.

  During step E104, a test is made to define whether the maximum bit rate reaches R ,,, ax (i) for the current picture (i = 0) is greater than the target bit rate R (i).

  If yes, go to step E107, otherwise go to step E105.

  Step E105 is a test to determine if the value of i is 0.

  If so, proceed to step E106 which sets the quantization step for the next image (image number 1) Q; + i to Qmin Thus, assuming two images successive are similar, the rate for the next image will be increased compared to that of the previous image.

  In the negative (i 0), step E106 is run to go to step E110.

  If the result of the test performed in step E104 is negative, then, in step E107, the bit rate / distortion allocation specific to the JPEG2000 standard is used, in the example considered, to reach the target bit rate Rc ( i).

  This step is identical to step El 1 of FIG.

  In the next step E108 a test is made to determine if the value of i is 0.

  If so, in the next step E109, the quantization step for the next image (image number 1) Qi + 1 is set to the value QmaxE Thus, the bit rate for the next image will be decreased compared to that of the previous image.

  In the negative (i 0) step E109 is run to go to step E110.

  The next step E110 which also follows the steps E105 and E106 is a step of incrementing the image number i in the sequence.

  In the next step E111, a test is performed on the value of i to determine if i is less than 2.

  If so, the steps E102, E103 and E104 already described are carried out again.

  2862168 27 If, on the contrary, i is equal to 2, then we go on to the next step E112 which initializes the number nmod (the number of images used to estimate the model parameters) to the value 2.

  Step E112 is then followed by step E2 of the algorithm of FIG. 5 which determines the target rate for the next image of the sequence (image number 2) and the steps E3 to E13 described above. are then executed again for the processing of this new current image.

  The invention makes it possible to rapidly regulate the rate of a sequence of images. For this purpose, the values of the rate Rmax (i) are compared with the target rate Ro (i) (steps E8 and E10) and, depending on the result, the quantization step prediction model is reset when the model is no longer suitable (step E9) and / or a rate / distortion allocation in case of exceeding the target rate R, (i) by Rmax (i) (step El 1).

  Note that by determining the quantization step of an image with a model adapted to the image, only the necessary data of the image will be processed by the entropic encoder, which will alleviate the task of the compression system 220 of Figure 2.

  In addition, the rate / distortion allocation step can sometimes be avoided, which again relieves the CPU and speeds up the rate control process.

  The adjustment of the quantization step generally results in a better approximation of the wavelet coefficients than the rate / distortion allocation.

  FIG. 7 represents, for a video sequence known to those skilled in the art under the name "Foreman" and compressed in the MJ2K format (Motion JPEG2000), the bit rate variations per pixel bpp as a function of the image number of the sequence.

  It should be noted that the value of the quantization step used for the compression is constant throughout the sequence and that no rate / distortion allocation method is used (for each image, all the data are kept).

  2862168 28 There are thus significant variations in bit rate, in particular during scene changes between the images of the sequence: in fact, from the image 0 to the image 180, the sequence represents a close-up of a face, of the In image 180 in image 230 a camera movement is observed, and from image 230, the sequence represents an overview of a building site.

  For each of these portions of sequence, the images have different characteristics from the point of view of the compression and the flow rate obtained is different.

  If you want to achieve a target rate of 1.25 bpp, for example, you can use a rate / distortion allocation method for each image between the image 0 and the image 180, as well as for each image between However, this method is not optimal: unnecessary data is processed during the compression and the rate-distortion allocation is used for each image between 0 and 180 and 210 and 300. On the other hand, the image 180 to the image 210, we obtain a non-optimal use of the available flow.

  Using the invention, the model used and a possible rate-distortion allocation on the first images allow a rapid convergence of the model. An optimized quantization step is thus obtained after a few images: only the useful data are then processed during the compression and the allocation step is avoided for the following images.

  In addition, from image 180 to image 230, model resets provide an optimized quantization step and better utilization of the available data rate.

  This provides a significant time saving for controlling the flow rate of the sequence.

Claims (52)

  1. A method for regulating the bit rate of a sequence of digital images, characterized in that it comprises, for a current image of the sequence, the following steps: determination (E2) of a target bit rate Re (i) for the current image, - determining (E4) a quantization step Q; corresponding to the target bit rate Rc (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the bit rate; - compressing (E5) the current image comprising in particular a step quantizing the image with the quantization step Q; previously determined, comparison (El0) of the maximum flow rate Rmax (i) obtained for the current image thus compressed with the target flow rate R, (i) in order to determine the greater of the two flow rates, - depending on the result of the comparison deciding whether to delete digital data from the digital data constituting the current image in order to bring the value of the rate Rmax (i) as close as possible to the target rate Rc (i).   2. Method according to claim 1, characterized in that it comprises a step of deleting (E11) data when the maximum rate Rmax (i) is greater than the target rate Rc (i).   3. Method according to claim 1 or 2, characterized in that it comprises a step of comparing (E8) the proximity of the maximum rate Rmax (i) with the target rate Rc (i).   4. Method according to claim 3, characterized in that it comprises a decision step for a possible reset of the quantization step prediction model for the next image of the sequence, depending on the result of the proximity comparison of Rmax (i) and Rc (i) for the current image, the decision of reinitialization of the model intervening when the model is no longer adapted to the current image.   2862168 30 5. Method according to claim 3 or 4, characterized in that it comprises a step of resetting the quantization step prediction model for the next image of the sequence when the relative difference between Rmax (i) and Ro (i) ) exceeds a predetermined threshold.   6. Method according to claim 4 or 5, characterized in that the decision to reset the quantization step prediction model depends on the number of images that have been previously compressed since the last reset of the model.   7. Method according to claim 6, characterized in that it comprises a step of resetting the quantization step prediction model for the next image of the sequence when the relative difference between Rmax (i) and Ro (i) exceeds a predetermined threshold and that the number of images previously compressed since the last reset of the model reaches a predetermined minimum number N. 8. Method according to one of claims 1 to 7, characterized in that it comprises a step of updating (E3) the parameters of the quantization step prediction model prior to the step of determining the quantization step of the current image.   9. Method according to claim 8, characterized in that the step of updating the model parameters for the current image takes into account the pairs of values (Rmax (i), Q;) obtained during the compression of the images of the sequence preceding the current image.   10. Method according to claim 8, characterized in that the step of updating the model parameters for the current image takes into account the nmod pairs of values (Rmax (i), Q;) obtained during the compression. nmod images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, the reinitialization of the model occurring when the model is considered to be no longer suitable for a current image.   11. Method according to claim 10, characterized in that, when the number nmod of images of the sequence which have been compressed since the last reset of the model has reached a predetermined maximum number N ,,, 2862168 31 then the step of updating of the model parameters for each of the following current images of the sequence takes into account the NW pairs of values (Rrnax (i), Q;) obtained during the compression of the NW images preceding the current image considered.   12. Method according to claims 3 and 8, characterized in that, when the relative difference between Rmax (i) and Rc (i) exceeds a predetermined threshold for the current image of the sequence, the update step model parameters for the next image of the sequence corresponds to a step of reinitializing the model which takes into account the m couples of values (Rmax (i), Q;) obtained during the compression of the last m images of the sequence , the number m being determined by the minimum number of pairs of values Rmax (i), Q; necessary to determine the parameters of the model.   13. Method according to one of claims 1 to 12, characterized in that, when the prediction model of the quantization step of the current image does not provide a mathematical solution to the determination of the quantization step, then the step of quantization is determined by averaging quantization steps Q; images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, a reinitialization of the model intervening when it is no longer adapted to the current image.   14. Method according to one of claims 1 to 13, characterized in that the prediction model of the quantization step of the current image is approximated by the equation R = aQ "1 + bQ-2, R and Q designating respectively the bit rate and quantization step for the current picture and a and b being the model parameters.   15. Method according to one of claims 1 to 14, characterized in that the image sequence complies with the Motion JPEG2000 standard and the deletion of data involves a post-compression distortion rate allocation algorithm.   16. A method for regulating the bit rate of a sequence of digital images, characterized in that it comprises, for a current image of the sequence, the following steps: determination of a target bit rate Rc (i) for the current image, - determining (E4) a quantization step Q; corresponding to the target bit rate Rc (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the bit rate; - compressing (E5) the current image comprising in particular a step quantizing the image with the quantization step Q; previously determined, - comparison of the proximity of the maximum rate Rmax (i) obtained for the current image thus compressed with the target rate Rc (i), - depending on the result of the comparison, decision on a possible reset of the predicting the quantization step for the next image of the sequence, the decision to reinitialize the model occurring when the model is considered no longer suitable for the current image.   17. The method of claim 16, characterized in that it comprises a step of resetting the quantization step prediction model for the next image of the sequence when the relative difference between Rmax (i) and Rc (i) exceeds a predetermined threshold.   18. The method of claim 16 or 17, characterized in that the decision to reset the quantization step prediction model depends on the number of images that have been previously compressed since the last reset of the model.   19. The method of claim 18, characterized in that it comprises a step of resetting the quantization step prediction model for the next image of the sequence when the relative difference between Rmax (i) and Rc (i) exceeds a predetermined threshold and that the number of images previously compressed since the last reset of the model reaches a predetermined minimum number N. 20. Method according to one of claims 16 to 17, characterized in that it comprises a step of updating (E3) the parameters of the model 2862168 33 prediction of the quantization step prior to the step of determining the step quantization of the current image.   21. Method according to claim 20, characterized in that the step of updating the model parameters for the current image takes into account the pairs of values (Rmax (i), Q;) obtained during the compression of the images of the sequence preceding the current image.   22. Method according to claim 20, characterized in that the step of updating the model parameters for the current image takes into account the nmod pairs of values (Rmax (i), Q;) obtained during the compression. nmod images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, the reinitialization of the model occurring when the model is considered to be no longer suitable for a current image.   23. The method according to claim 22, characterized in that, when the number nmod of images of the sequence which have been compressed since the last reset of the model has reached a predetermined maximum number N ,,, then the step of setting day of the model parameters for each of the following current images of the sequence takes into account the Nw pairs of values (Rmax (i), Q;) obtained during the compression of the Nw images preceding the current image considered.   24. The method as claimed in claim 20, characterized in that, when the relative difference between Rmax (i) and Ro (i) exceeds a predetermined threshold for the current image of the sequence, the step of updating the parameters of the model for the next image of the sequence corresponds to a step of resetting the model which takes into account the m pairs of values (Rmax (i), Q;) obtained during the compression of the last m compressed images of the sequence, the number m being determined by the minimum number of pairs of values (Rmax (i), Q,) necessary to determine the parameters of the model.   25. Method according to one of claims 16 to 24, characterized in that, when the prediction model of the quantization step of the current image does not provide a mathematical solution for the determination of the quantization step, then the step of quantization is determined by averaging quantization steps Q; images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, a reinitialization of the model intervening when it is no longer adapted to the current image.   26. Method according to one of claims 16 to 25, characterized in that the prediction model of the quantization step of the current image is approximated by the equation R = aQ "1 + bQ-2, R and Q designating respectively the bit rate and quantization step for the current picture and a and b being the model parameters.   27. Device for regulating the bit rate of a sequence of digital images, characterized in that it comprises, for a current image of the sequence: means for determining a target bit rate Rc (i) for the current image, means for determining a quantization step Q; corresponding to the target bit rate R, (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the bit rate, - means for compressing the current image including notably means for quantizing the image with the quantization step Q; previously determined, - means for comparing the maximum rate Rmax (i) obtained for the current image thus compressed with the target rate Rc (i) in order to determine the greater of the two rates, - decision means as to a possible deleting digital data from the digital data constituting the current image in order to bring the value of the rate Rmax (i) as close as possible to the target rate Rc (i), as a function of the result of the comparison.   28. Device according to claim 27, characterized in that it comprises means for deleting data.   2862168 35 29. Device according to claim 27 or 28, characterized in that it comprises means for comparing the proximity of the maximum rate Rmax (i) with the target rate Rc (i).   30. Device according to claim 29, characterized in that it comprises decision means for a possible reinitialization of the quantization step prediction model for the next image of the sequence, depending on the result of the proximity comparison of Rmax (i) and Rc (i) for the current image, the decision of reinitialization of the model intervening when the model is no longer adapted to the current image.   31. Device according to claim 29 or 30, characterized in that it comprises means for resetting the quantization step prediction model for the next image of the sequence.   32. Device according to claim 30 or 31, characterized in that the decision to reset the quantization step prediction model depends on the number of images that have been previously compressed since the last reset of the model.   33. Device according to one of claims 27 to 32, characterized in that it comprises means for updating the parameters of the quantization step prediction model.   34. Device according to claim 33, characterized in that the updating means of the model parameters for the current image take into account the pairs of values Rmax (I), Q;) obtained during the compression of the images of the sequence preceding the current image.   35. Device according to claim 33, characterized in that the means for updating the model parameters for the current image take into account the nmod pairs of values (Rmax (i), Q;) obtained during the compression of the nmod images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, the reinitialization of the model intervening when the model is considered no longer suitable for a current image.   36. Device according to one of claims 27 to 35, characterized in that the prediction model of the quantization step of the current image is approximated by the equation R = aQ-1 + bQ "2, R and Q denoting the rate and the quantization step for the current image respectively and a and b being the parameters of the model.   37. Device according to one of claims 27 to 36, characterized in that the image sequence complies with the Motion JPEG2000 standard and the deletion of data involves a post compression compression rate / distortion algorithm.   38. Device for regulating the bit rate of a sequence of digital images, characterized in that it comprises, for a current image of the sequence: means for determining a target bit rate Rc (i) for the image current, - means for determining a quantization step Q; corresponding to the target rate Rc (i) for the current image, the determination of the quantization step involving a mathematical model for predicting the quantization step as a function of the flow rate, - means for compressing the current image including means for quantizing the image with the quantization step Q; previously determined, - means for comparing the proximity of the maximum rate Rmax (i) obtained for the current image thus compressed with the target rate Rc (i), - decision means for a possible reset of the prediction model of the no quantization for the next image of the sequence, depending on the result of the comparison, the decision to reset the model involved when the model is considered no longer suitable for the current image.   39. Device according to claim 38, characterized in that it comprises means for resetting the quantization step prediction model for the next image of the sequence.   40. Device according to claim 38 or 39, characterized in that the decision to reset the quantization step prediction model 2862168 37 depends on the number of images that have been previously compressed since the last reset of the model.   41. Device according to one of claims 38 to 40, characterized in that it comprises means for updating the parameters of the quantization step prediction model.   42. Device according to claim 41, characterized in that the updating means of the model parameters for the current image take into account the pairs of values (Rmax (i), Q) obtained during the compression of the images of the sequence preceding the current image.   43. Device according to claim 41, characterized in that the means for updating the model parameters for the current image take into account the nmod pairs of values (Rmax (i), Q;) obtained during the compression of the nmod images of the sequence preceding the current image and which have been compressed since the last reinitialization of the model, the reinitialization of the model intervening when the model is considered no longer suitable for a current image.   44. Device according to one of claims 38 to 43, characterized in that the model for predicting the quantization step of the current image is approximated by the equation R = aQ-1 + bQ-2, where R and Q designate respectively the bit rate and quantization step for the current picture and a and b being the model parameters.   45. Communication apparatus comprising a device according to one of claims 27 to 37.   46. Communication apparatus comprising a device according to one of claims 38 to 44.   47. A computer-readable or microprocessor-readable information storage means having code instructions of a computer program for performing the steps of the method according to one of claims 1 to 15.   48. A computer-readable or microprocessor-readable information storage means having code instructions of a computer program for carrying out the steps of the method according to one of claims 16 to 26.   49. Removable or partially or completely removable computer-readable information storage medium comprising code instructions of a computer program for carrying out the steps of the method according to one of claims 1 to 15. .   50. Removable or partially computer-readable information storage medium or microprocessor comprising code instructions of a computer program for executing the steps   method according to one of claims 16 to 26.   51. Computer program loadable in a programmable apparatus, comprising sequences of instructions or portions of software code for implementing steps of the method according to one of claims 1 to 15, when said computer program is loaded and executed by the programmable device.   52. Computer program loadable in a programmable apparatus, comprising sequences of instructions or portions of software code for implementing steps of the method according to one of claims 16 to 26, when said computer program is loaded and executed by the programmable device.