FR2878384A1 - VIDEO COMPRESSION BY MODIFICATION OF QUANTIFICATION BY ZONES OF IMAGES - Google Patents

Abstract

The invention relates to the field of video signal processing, and more particularly to a video signal coding method adapted for compression. It relates to a method of encoding an original video signal comprising a temporal succession of images comprising quantized image data in the spatial domain, for obtaining a modified video signal, characterized in that it comprises the steps of: - receiving said original video signal, - cutting the images of said video signal into a set of image areas, - defining a new quantization of the image data in the spatial domain, variable according to the zones of said set of zones, for obtaining a modified video signal.

Description

VIDEO COMPRESSION BY MODIFICATION OF QUANTIFICATION

  The present invention relates to the field of video signal processing systems, these systems can be transmitters (camera, encoders), receivers (decoders, screens) nodes transmission, or storage where can perform a signal transformation, such as transcoding.

  Video compression systems aim to reduce the volume of the video signal in order to transmit it or to store it. The volume of the original video signal depends on several parameters. In the first place, these parameters are those relating to the digitization of the video signal. The digital video signal is defined as a temporal succession of images, each image consisting of pixels, each pixel being encoded on a number of bits, this number of bits per pixel being known as color depth. A pixel of a color video signal is defined by its three color components that are expressed in a given color space (the RGB color space, the YUV color space, etc.). Thus, the parameters relating to the digitization of the video signal are the image frequency, the spatial resolution of the image in pixels and the number of bits describing the pixel on each of its color components.

  With the advent of access to audiovisual data through very low-speed networks, such as mobile networks, it is essential to provide techniques to further compress the video signal. The prohibitive cost of deployment of these networks, coupled with a limitation of performance, especially in overload situations, makes it imperative. On the other hand, it should be noted that the terminals to display these video signals are increasingly disparate in their screen characteristics.

  In the compression standards in force, the parameters mentioned are taken into account by the standard only statically. Thus, to the stream result of the compression of a video signal are added information giving the value of these parameters. The objective is to allow the decoder to know this information to be able to configure itself.

  The current compression techniques do indeed provide for modifying the quantization of the video signal, but this modification takes place in the frequency domain, on the DCT (Discrete Cosine Transform) coefficients of the signal. In this area, the data is spatially decorrelated and is no longer related to the major characteristic color depth parameter of the screen to display the video signal. Thus, the quantization of the DCT coefficients is less effective than it could be since it does not take into account the strong spatial correlation or the application context that represents the screen of the terminal.

  On the contrary, the method according to the invention takes advantage of the spatial correlation as well as takes advantage of the temporal correlation.

  In addition, the known MPEG-4 ISO-IEC 14496-2 compression standard defines tools for indicating that a portion of a given image has been compressed with a spatial resolution reduced by half in each of the horizontal and vertical directions. compared to the resolution of the video sequence.

  To do this, the MPEG standard plans to analyze the input signal to identify the VOP (Video Object Planes) corresponding to specific objects in the image, and change the spatial resolution associated with certain VOPs. The standard provides for a uniform modification of the YUV components.

  More particularly, PCT application WO 03/107678 discloses non-uniform modification of these components and associated syntax.

  This compression process has two major disadvantages. On the one hand, the VOPs as defined in MPEG-4 are obtained semiautomatically, and the definition of a VOP uses an operator. The modification of the spatial resolution according to the VOPs therefore requires this intervention which is not compatible with an efficient compression method. In addition, it is possible that the modification of the spatial resolution according to these VOPs is not the most suitable modification for the compression.

  The present invention therefore intends to solve the disadvantages of the prior art as defined above. To this end, the present invention relates to a method of encoding an original video signal comprising a temporal succession of images comprising quantized image data in the spatial domain, for obtaining a modified video signal. , characterized in that it comprises the steps of: - receiving said original video signal, - cutting the images of said video signal into a set of image areas, - defining a new quantization of the image data in the spatial domain, variable according to the zones of said set of zones, for obtaining a modified video signal.

  Preferably, said cutting of the images is variable according to the number of the image in said temporal succession of images, which makes it possible to adapt the quantization to the image succession.

  Advantageously, said new quantization is variable according to the number of the image in said temporal succession of images, and this, whether the cutting remains identical or not, again to adapt the compression temporally.

  Preferably, the method according to the invention further comprises a step of analyzing said original video signal, and said division and / or said new quantification depend on said analysis. In this way, the selected areas and the associated quantization are optimized for compression. According to one variant, the analysis of the images concerns at least the spatial proximity, the movement characteristics and the residual energy characteristics of the image blocks.

  According to another variant, said division and / or said new quantization are predetermined.

  According to yet another variant of the method, said division and / or said new quantization depend on an action external to the method, such as, for example, the characteristics of the receiving terminal.

  According to one embodiment, the new quantization concerns the spatial resolution according to the image components. In this case, said images comprise a set of pixel-sampled digital image data within said image, and said new quantization is a resampling of the spatial resolution according to at least one of said digital image data, the data digital images being for example the color components.

  According to another embodiment, the new quantization relates to the number of bits on which the image data are coded, and in this case, said images comprise a set of digital data sampled in pixels, and are quantized over a number of bits. of origin, and said new quantization is a new quantification of said number of bits according to at least one of said digital image data or the manner of quantifying the image data on these bits, the digital image data being for example the color components.

  More generally, by combining the two embodiments above, said images comprise a set of digital data sampled in pixels, and quantized over a number of original bits, and said new quantization is a new quantization of said number of bits according to at least one of said digital image data or the manner of quantizing the image data on these bits, as well as a resampling of the spatial resolution according to at least one of said digital image data, for example the color components .

  According to one variant, said images of the original video signal are cut into a plurality of predefined size blocks, and in that said image areas obtained during the cutting step correspond to a plurality of said adjacent blocks.

  The invention also relates to an apparatus for encoding an original video signal comprising a temporal succession of images comprising space-domain quantized image data, for obtaining a modified video signal, characterized in that it comprises means for receiving said original video signal, means for splitting the images of said video signal into a set of image areas, means for generating a new quantization of the image data in the spatial domain, variable according to the zones of said set of zones, for obtaining a modified video signal.

  The invention also relates to a recording medium on which a series of images of a video scene coded according to the method according to the invention are stored.

  It should be noted that the process described in the present invention is compatible with the compression standards in force.

  The invention will be better understood on the basis of the description, given purely for explanatory purposes, of one embodiment of the invention, with reference to the appended figures: FIG. 1 generally illustrates the description of FIG. a video signal in images, FIG. 2a is an example of a division of a succession of images in image zones in which quantization varies, FIG. 2b represents a VOP slicing as defined in the prior art, FIG. 3 is a diagram illustrating the implementation of the invention in an encoder, FIG. 4 details the compression entity described in FIG. 3,

  - Figure 5 gives a more detailed description

  of the invention according to a possible embodiment in a video coder using the compression scheme based on the spatial prediction and the temporal prediction,

  Figure 6 gives a description of the process of

  quantization modification according to the invention, FIGS. 7a and 7b are examples of algorithms for defining the zones according to the invention.

  In addition, Table 1 describes the variables taken into account in an exemplary algorithm implemented by the present invention.

  The subject of the invention is a method for modifying the quantization of the color components of a video signal in order to improve the video compression / decompression systems and more generally the video signal processing systems.

  The video signal is described in Figure 1. This figure gives a structure of a video sequence as commonly used today. The video sequence, of structure as described in the figure, consists of a periodic succession of images Ij of the same size HxL. In this representation, j indicates the time reference. An image is composed of pixels, the number of which defines the spatial resolution of the image. Thus, an image at a time j is an array of HxL pixels. This example does not exclude other types of video, and in general it should be noted that a video signal can be described as a temporal succession of two-dimensional or three-dimensional images. The periodicity of the temporal succession, as well as the geometry and the image size can be variable. These images can result from a natural shooting through a sensor such as a camera, or a computer program for image synthesis, or drawings, or any other material and / or software and / or human. These images can equally be formed of a combination of images generated separately and combined by any editing process.

  As in any color system, a color dot is represented by three components: the RGB components for example. Generally in a video compression system, three other components, Y, U and V, which are a combination of the RGB components, are used. Other triplets of components than those mentioned above are defined and can be used just as much, these triplets of components being linked to each other by a linear combination. Whatever the selected triplet of components, we will now note in this document the color components C1, C2 and C3. In the exemplary image structure considered in Figure 1, we consider the simplest, namely that each pixel PX, y of the image has a value given by a triplet C1, C2 and C3. In conventional video systems, the three color components C1, C2 and C3 are quantized over a generally identical number of bits. Quantification is usually linear.

  In the present invention, we propose to modify the quantization of the signals representing each of the three color components C1, C2 and C3. For this, we define the notion of zone on which the modification is carried out. As drawn in FIG. 2, an area is a set of pixels of the video image, the pixels possibly being neighbors or not in the same image. According to this definition, the object of the invention is to modify the quantization of C1, C2 and C3 for a group of zones Zk into a predefined quantization Q1, for a given duration of images, the duration of images possibly being go from an image to the number of images constituting the total video sequence. Thus, by the method described here, over the duration of the video sequence, a zone Zk will have its components quantified chronologically according to, for example, the type Q1 during 100 images, then Qm during 1000 images, then Qn during 50 images etc. It is the object of this invention that the zones can be made and discarded as and when advancement in the video sequence.

  It is also noted that according to another embodiment of the invention, the modification of the quantization relates to the spatial resolution of the original video signal. In this case, the spatial resolution is modified according to at least one of the color components. As previously, this modification can be variable according to the zones obtained by the method of the invention, and as a function of the image or the image group chosen, for a temporal variation. In this way, a spatio-temporal variation of the spatial resolution of the video signal is obtained.

  Those skilled in the art will also understand that the above two embodiments can be combined by modifying both the spatial resolution of the image per zone, and the number of bits on which the color components or the color component are quantized. way of quantifying the color components on these bits.

  In general terms, for purposes of the present application, changes in the number of bits on which the color components are quantized or in the manner of quantizing the image data on these bits, which corresponds to a quantization in depth, the modifications relating to the spatial resolution of the image, which corresponds to a spatial quantization as well as the modifications on these two types of quantization in a temporal manner, therefore according to the index of the image.

  It will be understood that throughout the remainder of the present description, the quantization modifications applied to the number of quantization bits are also applicable to a modification of the spatial resolution, or to a combination of these two quantization modifications.

  According to the invention, the formation of the zones, the choice to modify the quantification thereof as well as the choice of the type of quantification are done by means of a control entity.

  This control entity can be based on different tools.

  These tools allow the analysis of the video signal which consists of studying each of the components forming the video signal to be modified, in a global manner-that is to say on the image-or local-that is to say on an area of the image. This analysis of the video signal calculates its complexity, namely its texture, likewise this analysis relates to its range, namely the set of values taken by each of the components, likewise this analysis relates to its cleanliness, therefore to the power of the noise included in the signal, likewise this analysis relates to its variation, in particular temporal, that is to say from one image to the following ones. This analysis step will be described in more detail later.

  On the other hand, the quality of service sought, particularly in terms of speed, average or instantaneous, is a decision element.

  On the other hand, the choice made by existing decision entities in the current compression standards, in particular on the use of temporal prediction or spatial prediction, and on their type, can be taken into account.

  Finally, a manual setting also helps the control entity.

  Thus, depending on the application, the entity may use all or part of these tools. Other decision criteria can be used as well.

  Finally, it should be pointed out that according to the method described the modified signal preserves the space-time space of the original signal.

  Figure 1 depicts a structure of the video as a temporal succession of images (j being the time index). Each image is an array of HxL pixels (along x and y axes), each pixel PX, y of coordinates (x, y) can be described at most by three values C1, y (j), C2X, y (j) C3x.y (j) Figure 2 describes the invention giving an example of image areas Zk. In this same example, the control entity makes the following decisions with regard to the area-based quantifications that are the subject of the invention: Quantification Q1 applied to Z1 and Z4 Quantification Q3 applied to Z2, Z5 and Z6 Quantification Q4 applied to Z3 and Z, Q1: C1, C2 and C3 quantized linearly over 8 bits Q3 Q4: Qu quantized C3 quantified C3: quantized quantized C3 C3 quantized linearly on linearly on nonlinearly nonlinearly nonlinearly 6 bits bit bits on 6 bits on 5 bits on 5 bits etc. In FIG. 3, a general view is given of the example of an embodiment of the invention, in which the video signal is caused to be compressed by an encoder.

  In this mode, the signal is received by the control entity which analyzes the signal to generate technical elements such as its complexity, the range of values taken by each component, its cleanliness, its time variation, etc. This analysis can be done after the Pre-treatment process.

  The same original video signal composed by C1, C2 and C3 is transformed by the preprocessing entity. The output of this entity is also a video signal described by three color component signals. These signals are compressed by the Compression entity that generates a signal that is called a bitstream. As will be detailed in FIG. 4, this compression entity may comprise functions of MPEG or H.26x compression standards, such as spatial prediction, temporal prediction, transformation (DCT, DWT, etc.), quantization coefficients, entropy coding, to name only the main ones.

  The control entity, by its analysis of the original video signal and / or other parameters, calculates zones Zk (see Figure 2) in the successive images. On some of these areas, the control entity reduces the quantization of each of the components C1, C2 and C3. In this embodiment, the reduction consists in decreasing the number of bits describing C1, C2 and C3 and thus moving to a number Ni, N2 and N3 (for each component) less than 8, 8 being the quantification assumed in the signal original video. The quantization modification takes place on signals representing each of the components C1, C2 and C3, as much in the preprocessing entity as in that of compression.

  The control entity adds in the outgoing bitstream of the Compression entity additional information such as those describing the zones and their quantization, as well as scene change information. It is understood that this additional information can be conveyed differently from the sender to the receiver, and they may not be routed, depending on the application. Other information can be used just as much.

  The raster bit stream enriched or not this information is then transmitted or stored.

  In FIG. 5, a more detailed example of an embodiment in which the video signal is made to be compressed by an encoder is given. In this embodiment, the quantization modification is performed in the compression entity. Still in this example, the compression entity is assumed to be based on a schema as detailed in Figure 4. In Figure 5, only the main functions and only the main signals are shown. Thus, for example and to name but a few, the control signals, the additional information signals (including in particular the choice of quantization modification, the zones formed in the image), the signal of the motion vectors, the storage entities (Buf Im) are not shown.

  In what follows, we will no longer use the terms of color components Cl, C2, C3 in the different processing phases, but simply the signal instead. The signal (1) of the original video passes through the Preprocessing entity (A) and becomes the signal (2) to be compressed. This signal is processed by the Spatial Prediction entity (B) and the Temporal Prediction entity (F).

  The Spatial Prediction entity (B) generates three signals: the signal (3) Image, the signal (10) Spatial Prediction and the signal (4) Residual Spatial. The signal Image is the set of pixels of the original image transmitted to the other compression steps (DCT, DWT or other type of equivalent transformation, quantization, reorganization, entropy encoding). The signal Spatial Prediction is the set of pixels considered as good prediction for spatially adjacent pixels and used as such. The Residual Spatial signal is the set of pixels represented by the difference between their original value and the value of the corresponding Prediction pixel. The signals (3) and (4) are then processed by the Quantization Modification entity (C) which is the subject of the present invention. The two resulting signals (5) and (6) are respectively the Image signal and the Spatial Residual signal after Modification of Quantization. These two signals are then transmitted to the following compression steps (DCT, DWT or other type of equivalent transformation, quantization, reorganization, entropy encoding). These same signals (5) and (6) are also processed by the entity (D) Scaling another object of this invention. The purpose of this entity is to operate the inverse function of the quantization modification entity (C). The resulting signals (7) and (8) are respectively the Image and Spatial Residual signal after resizing and are directed to the Reconstruction entity (E) which likewise receives the signal (10). The objective of (E) is to re-form the original pixels, in particular for the residual pixels. For said residual pixels, this re-formation of the original pixels is based on a simple addition of the value of the residual pixel and the value of the corresponding prediction pixel. The Spatial Rebuilt signal (9) then serves as a reference signal for the Spatial Prediction (B) in the same way as the signal (2) Image. In the same way, this signal (8) serves as a reference signal for the temporal prediction (F).

  The Time Prediction (F) entity generates two signals: the Spatiotemporal Prediction (15) signal and the Spatio-Temporal Residual (11) signal. In a manner analogous to that previously, the SpatioTemporal Prediction signal is the set of pixels considered as good prediction for spatially and temporally adjacent pixels and used as such. The Spatio-Temporal Residual signal is the set of pixels represented by the difference between their original value and the value of the corresponding Prediction pixel. The signal (11) is then processed by the Quantification Modification entity (G) object of the present invention. The resulting signal (12) is therefore the SpatioTemporal residual signal after quantization modification. This signal is then transmitted to the following compression steps (DCT, DWT or other type of equivalent transformation, quantization, reorganization, entropy encoding). This same signal (12) is also processed by the entity (D) of Scaling. The purpose of this entity is to operate the inverse function of the quantization modification entity (G). The resulting signal (13) is the Spatio-Temporal Residual signal after Scaling and is directed to the Reconstruction entity (E) which likewise receives the Prediction signal (15). The Spatial Time Reconstructed signal (14) then serves as a reference signal for the Time Prediction (F) as well as the signal (2) Image. Likewise, this Reconstructed signal (14) serves as a reference signal for Spatial Prediction (B).

  In order to clarify the interface between the entities described in this figure and functions (DCT, Q, and DCT- mainly) of the non-apparent compression entity, note that the entity of Reconstruction (E) also receives signals (30) coming from the other functions of the compression, in particular the signals having traversed the path DCT, Q, Q- 'and DCT-1.

  With regard to the Quantification Modification function performed by the entities (C) and (G), let us quote as an example the rounding operation with rounding.

  Thus, we have Output = rounded value (Input / Q) where Input is the value of a pixel at the input of the entity (C) or (G) and Output is the value of this same pixel at the output, Q being a constant whose value varies under the control of a control entity such as that shown in FIG. 3. This quantification modification example can be used in the case where it is desired to reduce the quantification of the signals and thus to gain in compression.

  Another example of simple quantization modification can be implemented according to the invention by the use of a bit-to-bit mask, with for example the instruction: Output = Input & Mask_Q

Example:

  if Input = 10110110 and Mask Q = 11110000 then Output = 10110000 With regard to the Scaling function performed by the entity (D), let us quote, also by way of example, the multiplication operation. Thus, to use the above notations applied to the entity (D), Output = Input x Q where Input is the value of a pixel at the input of the entity (D) and Output is the value of this same pixel at the output, Q being a constant whose value varies under the control of a control entity such as that shown in FIG.

  As regards the reconstruction function performed by the entity (E), note that this entity can, in addition to the re-forming function of the original pixel as already described, include other useful functions but not indispensable. Thus, it is conceivable to include a standardization function whose processing depends on the application considered. In general, the standardization function can be described according to the formula: Px, y (j) = Pxm, Yn (ji) 1 m ', n', m, n = 0,1,2,; i = 0,1,2, where P represents the pixel after uniformization function, P 'represents the front pixel, and x, y are the spatial coordinates of the pixel, j is the temporal index of the pixel ( the index of the image to which it belongs), xm, yn and xm ,, yn. are the spatial coordinates of pixels, pixels taken into account in the uniformization function, jr and ji are the temporal indices of pixels, pixels taken into account in the uniformization function, C is a linear combination of all or part of pixels.

  Figure 6 gives a description of the Quantification Modification phase and its surrounding elements. In this figure, a more general case than that described in FIG. 5 is considered. In FIG. 6, it is considered that the quantization modification can be done on the color component signals traversing the Pretreatment entity as appearing in FIG. FIG. 3, just as it can be done on the input color component signals of the two spatial and temporal prediction entities, or it can be done on the signals to be processed by the following compression steps (DCT, DWT or other type of equivalent transformation, quantization, reorganization, entropy coding) as described in Figure 5, etc. In FIG. 6, the Quantization Editing Signal Selection and Selection (Q) entity receives three signals: a signal (16), this signal being generally the signal Image 2) according to FIG. (17) on which the quantization modification will be performed and the signal (20) carrying analysis results performed or additional commands which will be described in more detail in the algorithm examples applied to the modification of the quantization. Entity (J) is the entity that performs Quantification Modification. This entity (J) therefore receives the signal (17) already mentioned and the signal (22) carrying the choice of quantization changes from the entity (H). The entity (K) is a simple Multiplexer, controlled through the signal (21) from the entity (H). Thus, the entity (K) will switch to the output (signal (19)) either the modified quantization signal (18) or the original signal (17). This switching takes place at the pixel rate. At the output, therefore, there is the signal (19) to be processed by the following entities and the signal (22) carrying the quantization modification choices.

  In general, the quantization modification takes place at least on the residual signals of the pixels situated in the zones concerned. Thus, for spatially predicted areas, the pixels that predict the neighboring pixels may not be affected by the quantization change.

  On the other hand, it is not excluded that the signal (17) is the same as the signal (16). In the example detailed above and repeating FIG. 5, the signals (16) and (17) are distinct.

  Note that the information conveyed by the signal (22) is taken again to be inserted possibly in the bit stream at the output of the Screening entity as shown in FIG. 3, and this as part of the additional information such as as previously described, just as this signal (22) can be used by other entities of the compression.

  Note also that the entity (H) is a part of the Control entity shown in Figure 3.

  The analysis means which will lead to the decisions of zone formation in the image and quantification modification (ie, the quantification of the image, that is to say the resolution spatial, or quantization of the pixel), quantization modification on each of said zones, analysis means best suited to the signal, the application and the context.

  In general, the quantization modification takes place at least on the residual signals of the pixels situated in the zones concerned. Thus, for spatially predicted areas, the pixels that predict the neighboring pixels may not be affected by the quantization change.

  Illustrated Figures 7a and 7b, we will now give examples of algorithms for the implementation of the invention to complete the description for the skilled person. These algorithms are based on current compression standards such as MPEG and H.26x.

  In the compression systems described in the cited standards, the images are cut into blocks of predefined sizes. According to specific criteria, the coder decides for each block of the type of prediction noted Type Pred and mode noted Pred mode for this prediction that will be operated on.

  On the other hand, the motion estimator receives a video signal. It calculates for the blocks to be predicted temporally the error on the block between the pixels of the potential predictor blocks and the pixels of the current block. It then retains as block predictor the one that gives the minimum error, which will be noted Pred error. He deduces the corresponding motion vector, which will be noted MV.

  For spatially intra-predicted blocks, a prediction error that will also be noted Pred error is usually calculated to choose among the different spatial prediction modes.

  As described in the above embodiment, the signal after passing through a) i.e. by quantization modification and rescaling functions or by the channel b), that is to say by the quantization modification functions, DCT, quantization and its inverse function of the coefficients and rescaling, is reconstructed by the entity (E). The reconstructed signal along one of two channels a) or b) is compared to the original signal, and an error between the two signals is calculated. This error will be noted Error Posteriori.

  Thus, each block of an image at a given time index has a set of variables that are: Pred type, Pred mode, aPosteriori error, Pred and MV error, except intra blocks for which the MV variable does not exist. We define for Error a _Posteriori, Error Pred and MV thresholds respectively S EaPi, S Ej and S MVk. Pred type and Pred mode have meanwhile discrete values defined in the cited standards.

  Variable Block Variable Area Threshold Remark MV Average value of S_MVk The average is made by variables of the blocks in the ratio to the number of zone pixels included in the blocks in the zone Error_Pred Average value of SE./ The average is carried out by variables of blocks in the ratio to the number of zone pixels included in the blocks in the zone_a_Posteriori average value of S_EaPi The average is made by variables of the blocks in the ratio to the number of zone pixels included in the blocks in the zone Type_Pred Identical One supposes that the zones are only formed of blocks having the same Type_Pred Mode_Pred Identical It is assumed that the zones are formed only of blocks having the same Mode_Pred t Identical Temporal index

Table 1

  Table 1 above is a summary table of the variables of a block and their values taken for the zone formed of block (s). Other possibilities may be provided such replace the average values by sums, in which case the thresholds at which the values obtained by these sums will be compared will have to be multiplied by the factor corresponding to the number of pixels included in the blocks in the zone.

  Finally, in addition to these variables, external actions are considered, such as the characteristics of the screen of the target terminal, user configurations (for example the operator), a command issued by the debit controller entity, requests from the network for more or less debit and more generally context information of the application. For these external parameters, discrete values are defined.

  The diagrams in FIGS. 7a and 7b describe an example of zone formation algorithm. This algorithm is an automatic zone renewal algorithm in each image. The principle of this algorithm is that neighboring blocks spatially are gathered in a zone if their Error Preds and their MVs are close in values.

  For this, the blocks are scanned in the natural sense. A block may have as neighbors several areas formed of at least one block.

  Arrived at a given block, one forces the inclusion of the block in the zone which is on his left, or in default in the zone which is above, or one applies the algorithm described in what follows.

  If this block is a temporally predicted block, the following process (figure 7a) is applied: for each zone, we first check that the zone and the block have the same type and prediction mode. In the affirmative case, the difference between the vector MV of the current block and the vector MV of the zone is calculated, and the difference between the Pred error of the current block and the Pred error of the zone. We compare the difference of the errors with a threshold S E1: if the difference exceeds this threshold, we go to the next neighboring zone, otherwise we continue the process. The difference of the MVs is then compared with a first threshold S MV1, if this difference does not exceed the threshold, the block is included in the zone and the variables of this zone are updated by calculating the average values; if this difference exceeds the threshold, we compare the signs of MV of the zone and MV of the block: if the signs are different, we go to the next neighboring zone, otherwise we continue the process. We compare then the difference of the MVs with a threshold S MV2: if the difference exceeds this threshold, we go to the next neighboring zone, otherwise we include the block in the zone and we update the variables of this zone by calculating the values averages. Once a zone meeting the inclusion criteria has been found, it is possible to go to the next block or to test all the neighboring zones and then apply the option described in the following paragraph.

  In the case where the given block is not a temporally predicted block but a spatially predicted block, the algorithm part described in FIG. 7b is applied: the given block is included in any neighboring zone having the same type and the same prediction mode as this block, if the difference of Pred Prediction of the block and Pred error of the zone does not exceed S E2 in absolute value.

  An option not described in FIGS. 7a and 7b provides for selecting from the zones that meet the inclusion criteria of the processed block the zone for which the Pred error value is the lowest. To implement this option, for example, it will suffice during the processing of a block, to record the data relating to each area meeting the inclusion criteria and then to compare the values of the Pred errors and to make the right inclusion decision. If more than one zone meets this criterion, then these zones will only form a new zone including the processed block, and the variables of this zone will be updated by calculating the average values.

  Finally, any block that does not find a neighboring area that meets the inclusion criteria forms a new area.

  An example of a zone quantization modification algorithm is now described.

  We first recall that there are three types of information useful for this algorithm: the temporal information of the zone that is to say the index of the image to which it belongs, the variables calculated for a given area as given in Table 1 and external actions.

  For a given zone, we choose a new quantization according to the temporal index of the zone, the external actions and the result of comparison tests of the variables Pred type, Pred mode, MV and Pred error with the discrete values that the first two variables are likely to take or with the respective thresholds S MVk and S Ej for the last two variables. It is understood that the new quantization chosen may be the same as the original one.

  Once this first step is performed for a zone, we calculate Error a Posteriori that we compare with the thresholds S EaPi. Depending on the result of these tests, either we decide not to modify the choice of new quantification taken during the first step, or we decide not to change the original quantization, or else a new quantization is chosen. . In the latter case, we can then decide that the new choice is not definitive and decide to recalculate Error _Posteriori and compare it again with the thresholds S EaPi. This second step can be repeated a number of times.

  In an alternative to this example of zoning algorithm and to this example of modification algorithm for zone quantization, it may be decided not to apply a zone division and / or a quantization modification by zone if the number of successive images in which quantization modifications have been made exceeds a certain threshold or if the number of successive images in which quantization modifications have been performed on an area exceeds a certain threshold.

  Finally, in cases where variables indicated in the table are not usable, other algorithms may be provided, based on the same principles of spatial proximity and motion characteristics and residual energy.

  The decisions made by the algorithms described above are transmitted or stored for possible use by other functions of the compression.

  Those skilled in the art will finally understand that the method that is the subject of this invention as well as the various steps constituting it can be implemented in a hardware and / or software way, in any video signal processing system, video signal transmission or video signal storage. The invention relates more particularly to systems where the compression as well as the optimization of the video signal with respect to the screens of the terminals are of primary importance.

Claims (15)

  A method of encoding an original video signal comprising a temporal succession of images comprising quantized image data in the spatial domain, for obtaining a modified video signal, characterized in that it comprises the steps of: - receiving said original video signal, - cutting the images of said video signal into a set of image areas, - defining a new quantization of the image data in the spatial domain, variable according to the zones of said set of zones, for obtaining a modified video signal.   2. A method of encoding an original video signal according to claim 1, characterized in that said division of the images is variable according to the number of the image in said temporal succession of images.   3. A method of encoding an original video signal according to claim 1, characterized in that said new quantization is variable according to the number of the image in said temporal succession of images.   4. A method of coding an original video signal according to any one of claims 1 to 3, characterized in that it further comprises a step of analyzing said original video signal, and in that said division and / or said new quantification depend on said analysis.   5. A method of coding an original video signal according to any one of claims 1 to 4, characterized in that said cutting and / or said new quantization are predetermined.   6. A method of coding an original video signal according to any one of claims 1 to 5, characterized in that said division and / or said new quantification depends / depends on an action external to the process.   A method of encoding an original video signal according to claim 1, characterized in that said images comprise a set of pixel-sampled digital image data within said image, and that said new quantization is resampling the spatial resolution according to at least one of said digital image data.   A method of coding an original video signal according to claim 1, characterized in that said images comprise a set of digital data sampled in pixels, and quantized over a number of original bits, and in that said new quantization is a new quantization of said number of bits according to at least one of said digital image data or the manner of quantizing at least one of said digital image data on these bits.   9. A method of coding an original video signal according to claim 1, characterized in that said images comprise a set of digital data sampled in pixels, and quantized over a number of original bits, and in that said new quantization is a new quantization of said number of bits according to at least one of said digital image data or the manner of quantizing at least one of said digital image data on these bits, as well as a resampling of the spatial resolution according to at least one of said digital image data.   A method of encoding an original video signal according to any one of claims 7 to 9, characterized in that said set of digital image data corresponds to color components.   11. Coding method according to claim 1, characterized in that said step of cutting comprises the creation of a set or a plurality of sets of pixels in an image of said video signal, the pixels being defined by their spatial coordinates.   Coding method according to claim 1, characterized in that said images of the original video signal are cut into a plurality of predefined size blocks, and in that said image areas obtained during the cutting step. correspond to a plurality of said adjacent blocks.   13. Encoding method according to claim 12, characterized in that said blocks correspond to MPEG or H.26x blocks.   An apparatus for coding an original video signal comprising a temporal succession of images comprising quantized image data in the spatial domain, for obtaining a modified video signal, characterized in that it comprises means for receiving said original video signal, means for splitting the images of said video signal into a set of image areas, means for generating a new quantization of the image data in the spatial domain, variable according to the zones of said set of zones, for obtaining a modified video signal.   Recording medium on which a series of images of a video scene encoded according to the method according to any one of claims 1 to 13, are stored.