FR2880744A1 - PROGRESSIVE QUANTIFICATION METHOD AND DEVICE, DECODING METHOD AND DEVICE, COMPUTER PROGRAMS, SIGNALS AND CORRESPONDING DATA CARRIERS - Google Patents

Abstract

The invention relates to an iterative method of progressive quantization of coefficients, by successive quantization refinements comprising, after an initial quantization step (31) performed on each of said coefficients with a predetermined quantization step and delivering quantized data, in which a coefficient is said refined during a given iteration if said value assigned during said refinement step is non-zero, so that for each current iteration, at least until the penultimate iteration, and for each refined coefficient of said current iteration, during the following iteration: the classifying step assigns to said coefficient the type (34), the following stages of determination of a residue and refinement are not implemented.

Description

Method and device for progressive quantification, method and device

  decoding, computer programs, signals and corresponding data carriers.

  FIELD OF THE INVENTION The invention relates to the coding of digital data, and more particularly to the quantization of digital data, expressed for example in the form of a set of coefficients to be transmitted or stored. The invention also relates, of course, to the decoding and restitution of the data thus quantified.

  More precisely, the invention relates to progressive quantification, allowing a progressive reconstruction of the data, first of all in a basic form, then more and more precise, or refined. This technique is known in particular under the name of scalable compression. It allows a user to become acquainted with the data (in basic form) before all data is received and / or processed, and to adapt the required quality to the type of terminal and / or the needs of the user .

  The invention finds particular applications in the coding of video signals, and in particular for the coding of coefficients obtained by 3D wavelets. We can then obtain a compression of video sequences in scalable very effective form.

  Subsequently, we consider more particularly the context of a scalable video coder developed within the MPEG21-SVC development program. However, the present invention is not limited to this context of a video encoder alone, but to any scalable encoder, for example for the coding of audio signals, and more generally of all signals for which progressive quantization is of interest.

  2. Scalable video systems Currently, most video encoders generate a single compressed stream corresponding to the entire coded sequence. If several clients wish to use the compressed file for decoding and visualization, they will have to download (or stream) the complete compressed file.

  However, in a heterogeneous system (for example the Internet), not all clients have the same type of data access: the bandwidth, the processing capacities, the screens of the different clients can be very different (for example, on a single computer). Internet, one of the customers will be able to have an ADSL speed of 1024 kb / s and a powerful microcomputer while another will only benefit from a modem access and an electronic personal assistant (PDA)) .

  An obvious solution to this problem consists in generating several compressed streams corresponding to different bit rates / resolutions of the video sequence.

  This technique is known as simulcast. This solution is of course sub-optimal since the same information is encoded several times.

  More recently, so-called scalable video coding algorithms (adaptive quality and variable spatio-temporal resolution) have appeared for which the coder generates a compressed stream in several layers, each of its layers being nested in the layer of higher level.

  Such coders are very useful for all applications for which the generation of a single compressed stream, organized into several scalability layers, can serve several clients of different characteristics, and for example: - video-on-demand services (in English video on demand, or VOD), for terminals of the UMTS type, PC ADSL, TV ADSL ... mobility of session (resumption on a PDA of a video session started on a television, or on a mobile UMTS of a session started on the GPRS network); - continuity of session (bandwidth sharing with a new application); high definition television (single encoding to serve standard definition (SD) or high definition (HD) clients; videoconferencing (single encoding for UMTS clients and Internet clients).

  In this context, algorithms based on wavelet transformations have become essential. They are thus now being adopted by MPEG21.

  More precisely, the model that was recently adopted by MPEG-21 SVC (described in particular in the document "Scalable Video Model 3.0", by J. Reichel, M. Wien, H. Schwarz, ISO / IEC JTC 1 / SC 29 / WG 11 / N6716, October 2004, Palma de Mallorca, Spain) is based on a scalable coder strongly oriented towards AVC-type solutions. This new standard will be able to provide scalable medium grain streams in time, space, and quality.

  According to this approach, motion information and texture information are distinguished. In order to perform a rate adaptation, the texture information is coded using a progressive scheme: coding of a first level of minimum quality (called "Base Layer" in English, or Base layer) - coding of progressive refinement levels (called Enhancement Layer, or Layer of Enhancement).

  The most common technique for progressively encoding information is to perform bitmap coding. Another approach, retained in the SVM of MPEG21-SVC, is to perform successive quantifications on the signal.

  In a first step (n = 0), a signal S is quantized using a quantization step Q (0) and then the quantization information is coded. Its version reconstructed using quantization information is Rec (0).

  Then a series of iterations (n> 0) is performed. For this purpose, we have the reconstruction obtained in the previous quantization step: Rec (n-1). We then define a residue Res (n) = S-Rec (n-1).

  This residue is quantized by the quantization step Q (n) and then the quantization information is coded. The reconstructed version DR (n) of the residue signal is then added to the previous reconstruction Rec (n1) to define the new reconstructed signal: Rec (n) = Rec (n-1) + DR (n).

  The evolution of the quantization steps Q (n) can be free, but in a practical way, Q (n) = Q (n-1) / 2 is usually taken. The quantification of a value is done by the following formula: + sign (x) * f * Q Quantif: x H Quantif (x, Q)

Q

  where f E [o ,,} 4] is a parameter to define a dead-zone around the value 0, so as to favor the attribution of a value 0, and lj represents the operation of rounding to the nearest integer (that is, the integer operator), and sign (x) is 1 if x> 0, is -1 if x <0 and is 0 if x = 0.

  The inverse quantization is performed from the following formula: ## EQU1 ## FIG. 1 illustrates the principle of such a successive quantization coding scheme.

  Because of the number of limited values during a requantization (typically with a decrease of the quantization step by two), a refinement uses values -1, 0 or 1, and a coding for a signifier uses the values -1, 0 , 1.

  As for the bit plane coding, it is possible to cleverly group the coding information of the quantization refinement information into signifiance and on the way to refinement.

  3. Disadvantages of the Prior Art The solution currently used is a system of coding by successive quantization of the residues with a quantization step which is halved at each iteration. The use of the parameter f used to define a deadzone then induces faults in the quantization performed.

  It can thus be seen in FIG. 1 that, when refining a coefficient, reconstruction values are used which are linked to intervals greater than the intervals of the previous quantization. There is therefore no nesting of the quantization intervals. This results in a loss of efficiency: the value reconstructed for an interval is not optimal since it does not correspond to the most probable value. On the other hand, it can be seen in this same figure 1 that for certain coefficients certain values are not probable (for example refinement 11 in -1 in certain cases). The inventors have observed that these problems have two origins: the successive quantization intervals are not nested; Using the dead-zone parameter for quantization moves the reconstruction points that serve as the center for subsequent quantization and can make some states improbable.

  4. OBJECTIVES OF THE INVENTION The invention aims in particular to overcome these disadvantages of the prior art.

  More specifically, an object of the invention is to provide a technique of progressive coding of a coefficient, by successive quantization, which is more efficient, in terms of quality and / or throughput, than the known techniques.

  Another objective of the invention is to provide such a technique, which does not introduce any particular complexity or important treatments. In particular, an objective of the invention is to allow a simple and efficient decoding of the quantized data.

  A particular object of the invention is to provide such a technique, which is suitable for an encoder and decoder of the MPEG21-SVC type.

  5. Main features of the invention These objectives, as well as others which will appear more clearly later, are achieved by means of a method of progressive quantification of coefficients, by successive quantization refinements comprising, after a step initial quantization method performed on each of said coefficients with a predetermined quantization step and delivering quantized data, at least one iteration of the following steps: - classification of signifiance of the coefficients, affecting each of said coefficients a contextual information including a so-called non-significant type, when said quantized data is zero, and a so-called significant type; determining a residual, corresponding to a difference between a coefficient and a value reconstructed using the quantized data at the preceding quantization steps; refining, by quantizing said residue, with a quantization step lower than that used in the preceding quantization step, affecting one of the three values -1, 0 or 1, according to the invention, a coefficient is said to be refined when a given iteration if said value assigned during said refining step is non-zero. For each current iteration, at least until the penultimate iteration, and for each refined coefficient of said current iteration, at the following iteration: the classification step assigns to said coefficient the signifier type; the following steps of determining a residue and refining are not carried out.

  Advantageously, a refined type is assigned to a refined coefficient.

  Thus, according to the invention, a progressive quantization coding is performed by taking into account a contextual information item of Refined Signifier that makes it possible not to code refinement information successively to a pass where a coefficient has already been refined.

  As we will see later, this makes it possible to optimize coding, but also decoding. In particular, at the level of the decoding, the contextual information of Signifier Refined will be taken into account, which will make it possible not to have to read refinement information successively to a pass where a coefficient has already been refined.

  In a first embodiment of the invention, during the last iteration, no processing is performed. The coefficient therefore remains of the refined signifier type. According to a second embodiment, for each refined coefficient of the penultimate iteration given, the classification step also assigns to said coefficient the signifier type during the last iteration.

  According to an advantageous characteristic of the invention, said quantization steps put a procedure tending to favor the assignment of a value 0.

  In other words, we can create zones of attraction (dead-zone) in English) which attracts the value 0, less expensive in the coding. The implementation of this zone of attraction can be selective.

  Thus, according to a preferred embodiment of the invention, said 10 quantization steps use a quantization equation such as: Quantif: x H Quantif (x, Q) = lx + sign (x) * f * Q [ p-, where f E {0,%}, defines, if f = 0, or not, if f = 1/2, an attraction range at the value 0, Q is a quantization factor, x is a coefficient or a residue to be quantized, sign (x) is 1 if x> 0, is -1 if x <0 and is 0 if x = 0.

  Advantageously, said step of determining a residue comprises an inverse quantization step for determining said reconstructed value using the following equation: R (i), S (i) H i * (1 f Where R (i) is the residue of the iteration concerned, S (i) is significance information for the iteration concerned, i is -1, O or 1.

  Note that this value 2j is different from those conventionally used in progressive quantification.

  The invention also relates to a device for encoding digital coefficients, comprising progressive quantization means implementing the method described above.

  The invention also relates to a method for decoding progressively quantized coefficients by successive quantization refinements, comprising an initial inverse quantization step performed on coefficient reconstruction data with a predetermined quantization step and delivering a reconstructed coefficient and information. contextual, and at least one iteration of the following steps: inverse quantization of a residue, corresponding to a difference between a coefficient and a value reconstructed using complementary reconstruction data; updating said reconstructed coefficient, as a function of said residue, if it is non-zero.

  According to the invention, each iteration comprises a step of assigning contextual information to each coefficient, chosen from the following types: a type called non-significant when said reconstructed symbol is zero; a type called refined signifier, when said residue of the current step is non-zero; and - a type said signifier, otherwise.

  For each current iteration, at least until the penultimate iteration, and for each refined signifier type coefficient of said current iteration, at the next iteration: - the classification step assigns to said coefficient the signifying type ; the following steps of inverse residue quantization and updating are not implemented.

  Advantageously, when all the iterations have been performed, a selective shift is assigned to said reconstructed coefficient, said shift being a function of the type of said reconstructed coefficient.

  This improves the quality of reconstruction.

  The invention also relates to a device for decoding coefficients quantized progressively by successive quantization refinements, comprising inverse quantization means implementing the decoding method described above.

  The invention also relates to computer programs comprising program code instructions for performing the steps of the quantization and decoding methods described above, when executed by a microprocessor.

  The invention further relates to a quantized data signal according to a progressive coefficient quantization method described above, comprising data for first coefficient reconstruction and residue data for refining said reconstruction. Such a signal comprises contextual information for each coefficient, selected from the following types: a type called non-significant: a type called refined signifier; - a type said signifier.

  Finally, the invention relates to data carriers carrying at least one data signal as described above.

  6. List of Figures Other features and advantages of the invention will appear more clearly on reading the following description of a preferred embodiment of the invention, given as a simple illustrative and non-limiting example, and annexed drawings among which: FIG. 1, discussed in the preamble, illustrates the principle of successive quantization coding; FIG. 2 presents the general structure of a quantification device according to the invention; Figures 3 and 4 are state diagrams illustrating the allocation of contextual information according to whether f is 0 (Figure 3) or 1/2 (Figure 4); FIG. 5 is a simplified flowchart of the quantization method implemented by the device of FIG. 2; Fig. 6 is a flow chart showing the coding of a residue of Fig. 4; Figure 7 shows the general structure of a decoding device 5 according to the invention; FIG. 8 is a simplified flowchart of the decoding method implemented by the device of FIG. 7; Fig. 9 is a flow chart showing the inverse quantization of a residue of Fig. 8; FIG. 10 schematically represents the adapted reconstruction of the reconstruction levels adapted to the reconstruction levels following the progressive decoding of the coefficients.

  7. DESCRIPTION OF A PARTICULAR EMBODIMENT 7.1 GENERAL PRINCIPLES The present embodiment is intended for video coding, and more specifically for MPEG-21. It therefore aims to improve the technique described in the preamble, while relying on it.

  The invention therefore proposes a successive quantization technique with introduction of an additional context (meaning refined) in order to improve the performance of a progressive coding.

  As will be seen later, in addition to (and thanks to) the taking into account of this new context, the invention proposes the following specificities: use of adaptive nested quantization intervals and use of a dead-zone mechanism; and adapted reconstruction within the quantization intervals.

  The use of adaptive nested quantization intervals improves compression performance by limiting the allowed symbols to the minimum necessary.

  The use of a dead-zone makes it possible to increase the compression efficiency by favoring the appearance of zero coefficients.

  The use of a suitable reconstruction within the intervals makes it possible to take into account the non-uniform statistics within the quantization intervals and thus to improve the quality of reconstruction.

  According to a particular characteristic of the invention, provision is furthermore made for a displacement of the reconstruction value associated with a quantization interval by adding to the decoder a displacement value of the reconstructed value, this value of displacement being dependent on the state coefficient (NS (non-significant), S (significant), SR (refined)).

  7.2 quantification, or coding. progressive The encoder is based on successive quantization encoding. The basic structure of such an encoder is illustrated in FIG.

  This encoder comprises processing means 21, receiving coefficients to be encoded 22 (representative of the texture of the video signal), and delivering quantization data 23 for progressive inverse quantization. The coding method is stored in a program memory 24. A RAM 25 stores the intermediate data and the data to be transmitted or stored.

  In the present embodiment, the encoder uses the same quantization modes as those used in AVC (Advanced Video Coding (described in particular in ITU-T and ISO / IEC JTC1, "Advanced Video Coding for Generic Audiovisual Services", ITU -T Recommendation 11.264 ISO / IEC 14496-10 AVC, 2003)), and implemented in MPEG-21 SVC. This allows for maximum reuse of existing tools.

  Thus, the quantization at the encoder used is the following: x + sign (x) * f * Q Quantif: xi> Quant if (x, Q) = el, but with f E {o, 3.} At each iteration, we reconstruct the coefficient of the previous iteration, by inverse quantization, as follows: - in the first quantization step: InvQuantif: i H InvQuantif (i, Q) - (i + sign (i) * () - f )) * Q, +1 sii> 0 with sign (i) _ 0 if i = 0 -1sii <0 in successive quantization steps: R (dS (i) H i * (1 f) * 3Q Note that this equation is substantially different from that usually used The classical value used for the progressive quantization is i * Q, or even i * (1.5-f) * Q The progressive quantization thus obtained is illustrated in FIGS. = 0 and f = 1/2.

  Compared to the known approaches operating by passes of signifiance and refinement, one notes that in the pass of refinement, the refinement can be omitted (skip) If the type of the coefficient is SR (Meaning Refined).

  The evolution of the type of coefficient is defined by the state diagrams presented in Figures 3 and 4.

  There are three possible states (types): NS: not significant; S: signifier; SR: refined signifier.

  The introduction of the type SR (meaning refined) thus makes it possible to use nested quantization intervals and to have an efficiency in compression thanks to the definition of the necessary syntax according to the modes of the coefficients.

  In the case f = 0 (FIG. 3), we start from the initial quantization 31 to reach either the NS type (if the quantized coefficient considered is 0) or the SR type (if the quantized coefficient considered is non-zero). When the type is NS, and we code (32) another value 0, the type remains NS.

  On the other hand, if we code a new signifier (33), the type becomes SR. During the next iteration (except possibly if it is the last one), the type becomes S (jump operation, or skip 34).

  When the type is S, two situations are possible: if the refinement to code is 0 (35), the type remains S; if the refinement to be coded is 1 or -1 (36), the type becomes SR. The table below summarizes the different possible cases.

  e SR SR RI) RI) 9 S sll I s 4t) I s q.el s tt s t e s s e e r s s e r s s s s s s s s r e s s The approach is the same in the case where f = 1/2, as illustrated in FIG. 4, except that, starting from the initial quantization, one reaches the type S (instead of SR) if the quantified coefficient is non-zero.

  The table below summarizes the different possible cases.

  S i1 s s É1LLE_ É +1 R É1 R N

S SR SR SR

  S SR NS SR S S S SR SR SR S S

  A block diagram of the progressive encoding is illustrated in FIG. 5. An initial coefficient x 51 is processed. For this: step 52: the current level n is initialized to 0 and the quantization step used is set to a value Q0. This value is a parameter of the encoder to define the minimum rate (quality) of reconstruction; step 53: a quantification of the coefficient is carried out using the formula: x Q Quantitative: x H Quant (x, Q) + sign (x) * f * a - Q and the reconstructed value is then defined by: InvQuantif: i H InvQuantif (i, Q) = (i + sign (i) * (V2 f)) * Q where i is the quantization index.

  The quantization index is then sent to the entropy coding module. type

  Caced Type Caded Type The type of each coefficient is then defined as: - f = 0. If the quantization index is 0, the type of the coefficient is NS, otherwise, it is SR; - f = 1/2. If the quantization index is 0, the type of the coefficient is 5 NS, otherwise it is S. step 54: iterates the process: incrementation of 1 of n, and division by 2 of the quantization step step 55: the quantization of the residue, ie the difference between the original coefficient and the reconstructed coefficient, by the same quantization formula. The reconstructed value corresponding to the calculated refinement is then added to the reconstructed value of the coefficient; then, the next iteration (56) is performed up to a predetermined stopping criterion (number of iterations or quality level, for example).

  Figure 6 illustrates more precisely the quantification of the residue.

  The input information 61 is the original coefficient, the reconstructed coefficient and the type of the coefficient.

  The following operations are performed: If the type of the coefficient is SR (62), its new type is S (63) and then the next iteration (64) is passed. Otherwise the residue (65) is quantized; - The quantization 65 of the residue is done using the formula x + sign (x) * f * Q Quantif: x H Quantif (x, Q) = From the construction, the only possible values for the quantification index i are -1, 0 or 1; If the type of the coefficient is S (66), the symbol R (i) is then sent (67) to the entropic coder. Otherwise, we send (68) the symbol S (i) The type of the coefficient is updated: - The type of the coefficient is S: if i = 0 (69), the type becomes S (610), otherwise it becomes SR (611); The type of the coefficient is not S (therefore of type NS): if i = 0 (612), the type becomes NS (613), otherwise it becomes SR (614); The reconstructed coefficient is updated (615) by adding the quantization refinement: R (i) S (i) H> i * (1 f) * 3Q 2 - Next iteration 64: we increment n, and divide Q by two.

  7.3 decoder, or inverse inverse quantization The decoder functions as a decoder of a successive quantization encoding system, using the mode evolutions presented in the previous section.

  Figure 7 schematically illustrates such a decoder. This encoder comprises processing means 71, receiving coefficients to be decoded 72, and delivering reconstruction data 73 allowing progressive inverse quantization. The decoding method is stored in a program memory 74. A random access memory 75 stores the intermediate data and reconstruction data of the original signal.

  It will be noted that the evolutions according to the invention do not require, in this embodiment, a modification of the binary format of the generated streams, but only to adapt the encoding and decoding grammar by taking into account the new introduced SR context. . Indeed in the case of a context SR no information is written or read in the bit stream.

  The main steps in the reconstruction of a coefficient are illustrated in Figure 8.

  The information relating to the coefficient to be reconstructed 81 is introduced, then: One initializes i to 0, and the quantization step Q to Qo (82); Initial inverse quantization 83 is performed initially. According to the value of f, the types of coefficients are initialized: If, 0: if the reconstructed coefficient is zero, its type is NS, otherwise it is SR; If J1 / 2: if the reconstructed coefficient is zero, its type is NS, otherwise it is S; Proceed to the next step 84: n is incremented, and the quantization step Q is divided by 2; The inverse quantization 85 of the residue is performed and the reconstructed coefficient is updated 86. The steps of reverse residue quantization and update of the

  reconstructed coefficient are illustrated in FIG. 9: The system 91 is initialized with the reconstructed coefficient following the initial quantization mode, its type, as well as the current quantization step; If the type of the coefficient is SR (92), then its type becomes S (93), and the next iteration 94 is passed; Otherwise, one tests the type (95) to know if it is the type S or NS. For each respective case, a symbol R (i) (96) or S (i) (97) will be read; If the read symbol is zero (98, 910), the type is updated to s (911) or NS (912) respectively. Otherwise, the reconstructed coefficient is updated with the residue increment (913, 914) defined by: R (i) S (i) i> i * _ f) *, Q, the type becomes SR (915, 916) and proceed to the next iteration 94.

  For the next iteration 94, n is incremented, and the quantization step Q is divided by 2.

  According to an advantageous embodiment of the invention, and in order to improve the decoding reconstruction level, a refinement of the reconstruction values can be performed as illustrated in FIG. 10, by adding to the reconstructed coefficient: 0 if the reconstructed coefficient is no; - fd * sign (coeff) * Q1 if the reconstructed coefficient is non-zero and of type S (101); fd * sign (coeff) * Q2 if the reconstructed coefficient is non-zero and of type SR (102) where Q, represents the value of the quantization step used in the last decoding pass, and fd represents a parameter of positioning of the level of reconstruction. The parameter fd takes its values in the interval [0, 1].

  Figure 10 illustrates the reconstruction thus obtained, with f = 0 and fd = 0.5. 7.4 applications As already mentioned, the invention advantageously applies to video coding, in particular in the context of MPEG-21. In this context, the invention applies to coding and decoding, but also to the storage, transmission and reception of scalable signals, as well as to these signals themselves (characterized by the presence of the third contextual information) and data carriers (eg CDs and DVDs, magnetic media, memories, data servers, ...) carrying such signals.

  Moreover, the approach of the invention can be implemented for other types of progressive coding, and for example for audio signals. $ 1

Claims (13)

  A method of progressive quantization of coefficients, by successive quantization refinements comprising, after an initial quantization step (31; 41; 53) performed on each of said coefficients with a predetermined quantization step and delivering quantized data, at least one iteration (56) the following steps: - ranking of signifiance (55) coefficients, affecting each of said coefficients contextual information comprising a type said non-significant, when said quantized data are zero, and a type said significant; determining a residue (55), corresponding to a difference between a coefficient and a value reconstructed using the quantized data at the preceding quantization steps; - refining (55), by quantizing said residue, with a quantization step lower than that used in the preceding quantization step, affecting one of the three values -1, 0 or 1, characterized in that a coefficient is said to be refined during a given iteration if said value assigned during said refining step is non-zero, and in that for each current iteration, at least until the penultimate iteration, and for each refined coefficient of said iteration current, during the next iteration: the classification step assigns to said coefficient the significant type (34; 63); the following steps (65, 66, 67, 68, 69, 610, 611, 612, 613, 614, 615) of determining a residue and refining are not carried out.   2. Quantization method according to claim 1, characterized in that for each refined coefficient of said penultimate iteration, during the last iteration, the classification step also assigns to said coefficient the significant type.   3. Quantization method according to any one of claims 1 and 2, characterized in that said quantization steps (65) put a procedure tending to favor the assignment of a value 0.   4. Quantization method according to claim 3, characterized in that said quantization steps implement a quantization equation such that: Quantif: x a Quant if (x, Q) = x + sign (x) * f * Q   where f E {0,12}, defines, if f = 0, or not, if f = 1/2, a range of attraction to the value 0, Q is a quantization factor, x is a coefficient or a residual to quantify.   The quantization method according to claim 4, characterized in that said step of determining a residue (55) comprises an inverse quantization step, for determining said reconstructed value using the following equation: R (iS (i) ai * (1 f) * 3Q where R (i) is the residue of the iteration concerned, S (i) is signifiance information for the relevant iteration, i is -1, 0 or 1.   6. Device for progressively quantifying digital coefficients, by successive quantization refinements, comprising means for initial quantization of each of said coefficients with a predetermined quantization step, delivering quantized data, refinement means (21) performing at least one iteration of following steps: - classification of signifiance coefficients, affecting each of said coefficients contextual information comprising a type said non-significant, when said quantized data are zero, and a type said significant; determining a residual, corresponding to a difference between a coefficient and a value reconstructed using the quantized data at the preceding quantization steps; refining, by quantizing said residue, with a quantization step smaller than that used in the preceding quantization step, affecting one of the three values -1, 0 or 1, a coefficient being refined if said value is non-zero, characterized in said refinement means (21) associates with a coefficient refined coefficient information in a given iteration if the value assigned in said refinement step is non-zero and, for each current iteration, at least 1 the penultimate iteration, and for each refined coefficient of said current iteration, at the following iteration: assign to said coefficient the signifier type; do not carry out the following steps of determining a residue and refining.   A method of decoding progressively quantized coefficients by successive quantization refinements, comprising an initial inverse quantization step (83) performed on coefficient reconstruction data with a predetermined quantization step and outputting a reconstructed coefficient and contextual information. , and at least one iteration (87) of the following steps: inverse quantization of a residue (85), corresponding to a difference between a coefficient and a value reconstructed using complementary reconstruction data; updating said reconstructed coefficient (86) as a function of said residue if it is non-zero; characterized in that each iteration comprises a step of assigning contextual information to each coefficient, selected from the following types: a so-called non-significant type (912), when said reconstructed symbol is zero; a type called refined signifier (915), when said residue of the current step is non-zero; and 2880744 and a type said signifying (911, 93), otherwise, and in that, for each current iteration, at least until the penultimate iteration, and for each refined signifier type of said current iteration ( 92), during the following iteration: the classification step assigns to said coefficient the signifying type (93); the following steps (95, 96, 97, 98, 99, 910, 911, 912, 613, 914, 915, 916) of inverse quantization of a residue and updating are not carried out.   8. Decoding method according to claim 7, characterized in that, when all the iterations have been carried out, a selective shift (101, 102) is assigned to said reconstructed coefficient, said shift being a function of the type of said reconstructed coefficient.   Apparatus for decoding coefficients quantized progressively by successive quantization refinements, comprising initial inverse quantization means performed on coefficient reconstruction data with a predetermined quantization step and delivering a reconstructed coefficient and contextual information, and reconstruction means (71) performing at least one iteration of the following steps: inverse quantization of a residue, corresponding to a difference between a coefficient and a value reconstructed with the aid of complementary reconstruction data; updating said reconstructed coefficient, depending on said residue, if it is non-zero; characterized in that, at each iteration, said means (71) of reconstruction 25 implement a step of assigning contextual information to each coefficient, selected from the following types: - a so-called non-significant type, when said symbol rebuilt is zero; a type called refined signifier, when said residue of the current step is non-zero; and - a type said signifying, otherwise, 2880744 22 and in that, for each current iteration, at least until the penultimate iteration, and for each refined signifier type of said current iteration, when I ' following iteration, said reconstruction means: assigns to said coefficient the significant type; do not implement the following steps of inverse quantification of a residue and updating.   A computer program comprising program code instructions for performing the steps of the quantization method according to any one of claims 1 to 4 when said program is executed in or by a microprocessor.   A computer program comprising program code instructions for performing the steps of the decoding method according to any one of claims 7 and 8 when said program is executed by a microprocessor.   A quantized data signal according to a progressive coefficient quantization method according to any one of claims 1 to 4, comprising data enabling a first coefficient reconstruction and residue data for refining said reconstruction, characterized in that it includes contextual information for each coefficient, chosen from the following types: a type called non-significant: a type called refined signifier; - a type said signifier.   A data carrier carrying at least one quantized data signal according to a progressive coefficient quantization method according to any of claims 1 to 4, comprising data for first coefficient reconstruction and residue data for refinement. said reconstruction, characterized in that it comprises contextual information for each coefficient, selected from the following types: z3 a type called non-significant: a type called refined signifier; a type said signifier.