MXPA99004805A - Method and device for video compression - Google Patents

Abstract

The invention concerns a method for compressing digital data of a sequence of computer synthesised images describing a scene subject of a script, comprising a processing step (1, 11) for modelling the scene on the basis of mathematical data, a step of image rendering for creating a synthesised image from this modelling and storyboarding in blocks of images of this synthesised image (2, 12), a differential coding of the current image block from a block of at least one synthesised image defined on the basis of at least one motion vector for supplying a residual block (4i). The invention is characterised in that the motion vector is computed from mathematical data derived from the synthesised script and defining the apparent motion of the different objects constituting the scene subject of the sequence. The invention is useful for synthesising images in production. For video games, it is useful for interaction or for producing virtual reality.

Description

METHOD AND DEVICE FOR VIDEO COMPRESSION The synthesis of images makes it possible to create, with the help of computer tools, so-called virtual images. These emanate from an abstract description and digital calculations. This involves the use of a collection of methods via the two-dimensional and three-dimensional graphic libraries, which are possibly accelerated by the specific circuits of the type of accelerator graphic cards, and by means of suitable interconnections of the API type (meaning "Program Interconnection"). of application") . The process to create these images can be divided into several stages. This first comprises a modeling phase, that is, of computing or acquiring objects using a description model, the objective of which is to describe the constituent objects and • assemble them to construct an observable scene from them. For example, we can mention the model of the polygonal type in which the objects are divided into multitudes of elementary polygons or facets. Primitive graphics are used to define, assemble or modify these elementary geometrical entities. These models are interpretable: these can be associated with graphic engine types, for example colored (or "shaded") triangles, anti-pseudonyms of texture, etc. These have capabilities, that is, properties or capabilities, which are behavioral, such as movement, explosion, ..., and also visuals such as texture, color, mirror effects, etc. These will be able to interact with their environment when a scenario is built, for example with lights, with the other objects. There is, therefore, in the second place, a construction of a scene in motion, which governs the global organization of these models over time (taken here in the sense of time to effect a given application), that is, the definition of a stage or animation. In the end, depending on the applications (CAD, image production, simulation, etc.) the final step is to create digital images from this scenario. This last step is called interpretation or the method of "image interpretation"; the purpose of which is to interpret the scene as realistically as possible. This can be very expensive in terms of calculation time, and require large memory capacities with respect to the models used and with respect to the data related to the programs involved. For example, interpretation methods such as radiosity or ray processing make it possible to obtain quality images, but at an even higher cost, the algorithms of calculations implemented being very complex.

The volume of information represented by the digital images has given rise to the development of various compression standards such as JPEG, H.263, MPEG-1, MPEG-2, and soon MPEG-4 making it possible to manipulate, either for storage or the transmission, volumes of information that are compatible with the current technology. The MPEG-2 standard, now the most generic, makes it possible to compress all existing image formats with the help of various profiles and levels defined in the MPEG standard, the best known of which is MP @ ML ("Main Profile at the Main Level "), for images in the conventional television format. The structure of the encoders that perform such video image compressions, according to the prior art, rely on various types of images. Intra, Predicha or Bidirectional (I, P and B, respectively), the main difference being the temporal mode of prediction. The coding core is conventional with a frequency division based on the DCT ("Discrete Cosine Transformation"), followed by quantization and entropy coding, to obtain, at the output of the encoder, a binary stream that must comply with the standard, that is to say with a specific syntax. The temporal prediction is made by estimating the movement between separate images in time, based on the image blocks of size of 16x16 pixels for example. The movement is deduced from a correlation between the block of the current image and a block of a search window of a previous or subsequent image. Next, each 8x8 pixel size block of the image is predicted with the calculated displacement vector, and only the error between the estimate and the original is encoded. The compression of data, either conventional images or synthesis images, therefore uses conventional processes such as motion estimation. The circuits that carry out such calculations and the associated circuits are complex and the cost of such adjustment is high. For example, motion estimation and motion compensated interpolation circuits explain perhaps half the complexity of an MPEG-2 type encoder. Movement information, still according to conventional processes, does not always correspond to the effective movement. This simply involves the correlations in general with respect to the luminance information. The fact that the vector field consists of the motion vectors of an image does not reflect the effective movement excluding optimal compression of the data, in particular in the case of vector differential coding. This is because, for macroblocks corresponding to areas of uniform motion, the cost of identical or slightly different transmission vectors in differential coding is less than the cost of random transmission vectors. Furthermore, the fact that the motion vectors obtained according to the conventional "block equalization" process does not necessarily reflect the effective movement, excludes the use of the vector field to carry out the interpolations or extrapolations of the good quality images. for example during frequency conversions, slow motion modes of the digital video recorder, etc. A field of incorrect motion vectors also excludes the use of new coding techniques using the contour information for an image instead of the macroblocks. This is because the compression of the data according to these new techniques is based on the segmentation of images and the effective displacement of these "segments" that define the uniform zones. Thus, the lack of reliability of the motion estimation excludes the optimization of the operation of the encoder in terms of degree of compression or image quality for a given bit rate or rate or the effective use of this motion information in the decoder. The purpose of the invention is to solve the aforementioned drawbacks during the coding of the synthesis images.

For this purpose, its object is a process for the compression of digital data of a sequence of synthesis images, which describe a scene that is a scenario, which comprises a stage of processing for the modeling of the scene based on the mathematical data , a stage of image interpretation to create a synthesis image from this modeling, and a division of this synthesis image into image blocks, a differential coding of the "current image block based on a block of at least one synthesis image, this block being defined based on at least one motion vector, to provide a residual block, characterized in that the motion vector is calculated from the mathematical data that arise from the synthesis scenario and that define the apparent motion of the various objects that make up the scene, which is the objective of the sequence, its objective is also a device to compress the data s digital of a sequence of synthesis images, that describe a scene that is the objective of a scenario, that includes a processing circuit to model the scene, the images of which are going to be synthesized based on the mathematical data, a circuit for the interpretation of the image and to divide the image into blocks, which receives the reference points from the processing circuit to make a synthesis image, and divide the obtained image into image blocks, a motion compensation circuit of image blocks that receive the reference points from the processing circuit, to provide predicted blocks, a subtractor to take the difference between the current block that originates from the circuit for image interpretation, and to divide it into image blocks, and the predicted block that originates from the motion compensation circuit to provide a residual block, a transforming circuit discrete cosine for the image blocks originating from the circuit for the interpretation of the image, and for dividing it into image blocks or residual blocks originating from the subtracter, the choice being made by a selection circuit so as to function of the energy criteria, a circuit ^ to quantify the transformed coefficients, characterized in that the motion compensation circuit uses the mathematical data provided by the processing circuit and the representation of the displacement of the modeled objects constituting the scene, • to calculate the motion vectors associated with the current block and that define the predicted block. According to another modality more, its objective is a device for the compression of digital data of a sequence of synthesis images that describe a scene that is the objective of a scenario, which comprises a processing circuit for modeling the scene, the images "from which will be synthesized based on the mathematical data, a circuit for the interpretation of the image and to divide the image into blocks, which receives the reference points from the processing circuit, to make a synthesis image and dividing the obtained image into image blocks, an image block movement compensation circuit that receives the reference points from the processing circuit, characterized in that it transmits in intra mode an image of between N images of the sequence, being In a predetermined number, this image N is that which is the objective of the interpretation calculation by the interpretation calculation circuit. ion and to divide it into image blocks, and because the other images are transmitted in inter mode by means of residual blocks that represent the difference between a current block and a predicted block, and because the residual blocks are null and defined by the vector of simple movement calculated from the mathematical data. In general, the techniques of image interpretation come to represent a scenario oriented by "objects" as images. Now, the scene, that is, the stage, comprises all the possible information regarding the objectives in the scene, and also its various properties. In the case of picture synthesis, a two-dimensional or three-dimensional scene gives the exact displacement of objects over time. This scenario then serves to generate the final digital video images, final (interpretation). In this way, instead of using the information consisting of the pixels that constitute a visual image, that is, one that is not modeled, to estimate the movement, modeling tools are used to calculate the effective movement in the sequence of images. Apart from the reduction in complexity, through the use of effective movement instead of estimated movement, it is possible to improve the prediction quality and overall operation of the encoder. Other features and advantages of the invention will become clearly apparent in the following description, given by way of non-limiting example, and offered in conjunction with the accompanying figures which represent: - Figure 1, the architecture of a video compression device according to the invention; Figure 2, a simplified architecture of such a device. Figure 1 gives a description of a first version of the device according to the invention. As explained above, the calculations related to the synthesis of images are of great integral complexity carried out by dedicated work stations. It is such a station, referred to here as the processing circuit 1, which performs a modeling of a previously defined scene based on a scenario for the purpose of creating these synthesis images, representative of the scene that is the objective of the scenario. The information obtained in this way from the processing circuit is transmitted in parallel to an image "interpretation" circuit 2 and an interface or interconnection circuit 6. The interpretation circuit output is connected in parallel to the input of a circuit for dividing it into image blocks 3, and to a first input of motion compensation circuit 7. A first output of the division circuit is directly connected to a first input of a mode selection circuit 5. A second output i of between n of the division circuit is connected to a first input of a subtracter 4i of between n. The output of the subtractor 4i is connected to a second input i of between n of the mode selection circuit 5. The output of the interconnection circuit 6 is itself linked to a second input of the motion compensation circuit 7. An output i of between n of the last circuit is connected to a second input of the subtracter 4i. The output of the mode selection circuit 5 is connected to the output of the device through a Discrete Cosine Transformation calculation circuit 8, a quantization circuit 9, and a Variable Length Coding circuit 10, placed in series. The output of the variable length coding circuit, which is the output of the device, is also returned to a second input of the quantization circuit 9 (including the speed regulation or bit rate). Another output of the motion compensation circuit 7 is connected to a second input of the variable length coding circuit 10. In this example, the motion compensation 7 is performed without the previous decoded images, ie the reconstructed images. The problems of accumulation of prediction errors, a phenomenon that is known as "displacement" can then occur. To improve the process, it is possible to use a loop or negative feedback loop with inverse quantization, inverse DCT, which provides the compensation circuit 7 with the previous decoded images. The processing circuit 1 therefore symbolizes the computer tools required for the mathematical formulation of a scenario. Its function, as explained previously, is to model a scene in three dimensions, that is, to define the mathematical equations of the objectives that constitute the scene and their movements defined by the scenario. This mathematical modeling data can also originate from the computer files accessible to the processing and storage circuit, for example, the predefined models. The role of the image interpretation circuit is to create the synthesis images. This carries out the conversion in pixels from the modeled scene. The digital information of luminance and chrominance for the obtained image is transmitted to the division circuit 3 which carries out a division of each image in macroblocks, blocks of size of 16x16 pixels that comprise 4 blocks of image of 8x8 pixels, according to with the MPEG-2 standard The role of the interconnection circuit 6 is to convert the displacement information given by the application, here the image synthesis, in a movement field in the form of macroblocks which is transmitted to the motion compensation circuit 7. With each macroblock there is associated at least one motion vector calculated by this interconnection circuit from the mathematical data for the modeling of the scene, which are received from the processing circuit 1. The mathematical transformations such as translation, rotation , homotesia, etc. they are represented as two-dimensional vectors associated with each macroblock. Of course, the motion vectors represent the apparent movement of the objects, taking into account, for example, the displacement of the point of view of the scene. This conversion of the displacement is carried out as follows: let us assume that the displacement is an object translation defined by a three-dimensional vector whose projection in the plane of the image gives a vector of coordinates in the plane (Dx, Dy). With all the complete macroblocks that represent a part of this aspect, this same vector is associated which is considered as the motion vector (Dx, Dy), as defined in the MPEG standard. There will remain particular cases, such as macroblocks in the boundaries of objects, of covered or uncovered areas.

As much as the boundary blocks are related, the simplest solution is to take the majority vector in the macroblock, that is, the vector that corresponds to the largest number of pixels in the macroblock, as the motion vector associated with the macroblock. It is also conceivable to select the vector corresponding to the object instead of that of the antecedent. This choice is not of major importance since the device described has a mode selection circuit, which will choose the most suitable mode, for example, the intra mode if the reconstruction (and therefore the estimated movement) is not good. With respect to the covered / discovered macroblocks, the simplest solution is to assign the zero movement to the macroblock, knowing that the zones corresponding to these covered or discovered macroblocks will be poorly reconstructed. A more elaborate solution is provided by the MPEG-2 syntax, which makes provision for front and rear vectors with respect to the B (Bidirectional) images. These two types of vectors are then calculated by the motion interconnection circuit 6 based on the data provided by the processing circuit 1, and are therefore considered as front and rear movement vectors associated with the processed macroblock. Then, an uncovered area can be reconstructed with the movement directed forward (in time), a z.ona. covered by the movement directed backwards. The choice of the coding mode is tap made here by the mode selection circuit. The MPEG standard in fact makes it possible to assign various types of vectors to a macroblock. the monodireccipnal vectors "forward" or "backwards", which take into account respectively a previous image and an image subsequent to that corresponding to the processed macroblock; the bidirectional type vectors used in the case in which the mode selection circuit chooses bidirectional predictive mode. This in fact involves two vectors associated with the macroblock, the processed macroblock that is matched or compared, by a leading vector and a subsequent vector, with a macroblock that is one that averages the two macroblocks defined by these vectors, whose average is a function of the luminances of the matched or compared pixels and of the temporal distances to the macroblock. indicted . These vectors can be the same as those defined above. The motion compensation circuit 7 performs, based on the fields of motion vectors and the previous (and subsequent) images transmitted by the interpretation calculation circuit, the calculations of the predicted images, ie the compensated images in motion, as well as the division of the images obtained in this way in macroblocks. The current macroblock, transmitted by the division circuit 3 is received on the positive input of the subtracter 4, the corresponding macroblock of the predicted image, whose macroblock is transmitted by the compensation circuit 7, is received on the negative input of the subtracter. The "residue" available at its output, the macroblock corresponding to the difference formed with respect to each pixel, is transmitted to the mode selection circuit 5. In fact, with each type of motion vector defined above corresponds a predicted image calculated by the compensation circuit. This circuit therefore has as many outputs n as types of chosen vectors, and each output is connected to a subtracter 4i which receives the current block of the image on the other input. All calculated waste types are transmitted to the mode selection circuit on their n inputs, which receives, in another input, the current macroblock, directly from the division circuit 3. The mode selection circuit 5 determines the mode which the macroblock will be encoded. The available coding modes are, for example: encoding with motion compensation or without motion compensation, ie with or without transmission of the motion vector; intra or intra coding (inter-image coding, inter-frames of similar parity or non-similar parity, predictive or bi-directional predictive coding); if the coding is not intra, the coding of the waste or not (in the case where the waste is very small). The choice of mode is made by the mode selection circuit, by performing a macroblock energy calculation in each mode, and by choosing one that gives the smallest energy. In most cases, the calculation of energy is approximated by the sums of absolute values. The selected block then undergoes a Discrete Cosine Transformation, a quantization and a coding of the Variable Length Coding type. The proportion of bits is minimized at the output of the device to be transmitted to the quantizer, this loop or regulation loop acting on the quantization interval in the form of a block for regulating the speed or proportion of bits, included within the quantization circuit for simplification purposes. The vectors associated with the processed macroblock are transmitted by the motion compensation circuit to the variable length coding circuit 10, which performs a multiplexing of these values with the quantized coefficients of the macroblock. It is of course conceivable to simplify this device, for example by limiting the type of predicted macroblocks subjected to the subtractors. The choice of the mode could then be carried out par --.-.--- by the interconnection circuit, when the latter can deduce, from the information it possesses, the mode that allows the best correlation. The interconnection circuit then transmits a mode reference point on an input -supplementary circuit 5 mode selection, to impose this mode. The mode selection circuit is then a simplified circuit. Another illustrative embodiment is described in Figure 2. The output of a processing circuit 11 is connected in parallel to the input of an image interpretation calculation circuit 12, and to the input of a motion interface or interconnection circuit 13. A bidirectional connection connects these last two circuits. The output of the image interpretation calculation circuit 12 is connected to the output of the device through a cosine discrete transformation calculation circuit 14, a quantization circuit 15 and a Variable Length Coding circuit 16, placed in series . The output of the Variable Length Coding circuit, which is the output of the device, is also returned on a second input of the coding circuit 15 to constitute a loop or regulation loop. The output of the interconnection circuit 13 or movement interface is connected to a second input of the Variable Length Coding circuit 16. The processing circuit 11 has the same characteristics and functions as that described in Figure 1. The interpretation calculation performed by the circuit 12 is here different from that described in Figure 1, in the sense that it is made by the taking of samples, that is to say for an image every N images of a sequence of images that are going to be transmitted. The image constructed by this circuit is then divided into macroblocks and image blocks, whose function is also carried out by this image interpretation circuit. The blocks are transmitted in succession to the circuit for the calculation of the cosine discrete transformation 14, which carries out the cosine transformation of these blocks, towards the quantization circuit 15 which performs a quantification of the transformed coefficients, a delay and a serialization of these coefficients, and finally to the Variable Length Coding circuit 16 which carries out an entropy coding of the serialized data. The loop or loop of regulation decreases the proportion or speed of bits at the output of the device, to supply a bit rate regulation circuit, built in the quantizer 15, which acts on the quantization interval to perform such regulation. The motion interconnection circuit 13 calculates the fields of the motion vector in a manner analogous to that described for the motion interconnection circuit 6 of FIG. 1, and transmits the vector or motion vectors assigned to the current processed macroblock to the circuit of Variable Length Coding. It also provides the reference point corresponding to this or these vectors. The coding circuit performs a multiplexing of the motion vector associated with the current macroblock, together with the coding mode and the transformed and quantized data of this macroblock, to transmit all this as the output from the device. The connection or connection between the movement interconnection circuit 13 and the interpretation calculation circuit 12 makes it possible to exchange the information related to the division into macroblocks, the current macroblock transmitted to DCT, the coding mode, etc. The image interpretation procedure, which consumes a large amount of calculation time, is therefore simplified here. The idea is to use only certain synthetic images and movement between the images that are then missing. The calculated synthetic images are separated in time (i.e., one image per N image periods) and typically represent the images encoded in the intra mode by the encoder. The so-called missing images are neither calculated, transmitted nor stored. The decoder is responsible for the creation of these with the help of the reference point of the movement which is then transmitted to it. In this way, everything happens as if the encoder were in a configuration in which the prediction error images were null (the compensated prediction in motion is eliminated). The coding mode transmitted by the circuit 13 is dependent on the images processed by the interpretation circuit. With the calculated images corresponding to an intra coding, the motion vectors for these images are not transmitted. The intermediate images are considered as images encoded in the inter mode together with a zero prediction error (coefficients of the residual macroblock at zero), the movement represented by the motion vectors that are the effective calculated movement and the unconsidered movement, and the synthetic images assumed to be free of noise. In the decoder, these inter images are recreated from the intra image and from the transmitted motion vectors for each inter image, as indicated above. In this application, the movement in the form of macroblocks (motion vector) can be inaccurate, the defects in the limit blocks are not corrected and it is assumed that the illumination is constant between the intra images. Such a device is reserved for applications in which the image quality may be lower, such as video games. Processing speed, data compression and low construction cost are features imposed on such devices, to the detriment of the quality of the image. In a first improvement of this device, in terms of image quality, the coding mode depends on the type of macroblock processed. An error interconnect circuit 17, shown in broken lines in Figure 2 is added to carry out this new function. This circuit receives the information coming from the processing circuit 11 for the exchange, with the circuit 12 for calculating the image interpretation, the information related to the coding of the current macroblock. Its role is to detect errors that occur mainly in the contours of moving objects. The specific processing is provided for the relevant macroblocks. When the macroblocks to be transmitted are limit blocks, the information is either computed by the circuit to calculate the errors based on the information originating from the processing circuit 11, or received from the interconnection circuit 13 of motion via the image interpretation calculation circuit, the error interconnection circuit imposes, on the image interpretation calculation circuit, the calculation of the image that interprets these limit blocks. The mode is then forced into the intra mode for this macroblock, the information transmitted to the motion interconnection circuit 13. A simplified interpretation calculation is therefore performed by the image interpretation calculation circuit 12 for the images encoded in the inter mode, just for these blocks of the image. A second improvement of the device consists of adding a lighting interconnection 18, represented by dashed lines in Figure 2. This lighting interconnection 18 receives the information coming from the processing circuit 11 for interchange, with the image interpretation calculation circuit 12 , the information related to the coding of the current macroblock. The function of this circuit is to convert the lighting model into an error image for the encoder. The change from an object reference point to a suitable reference point for coding according to the MPEG standard is done, for example, when considering the change of illumination coming from one image to another because it is a prediction error. In this way, the reference point related to the variation in the luminance of the processed macroblock coming from one i-magen to another, is transmitted to the image interpretation calculation circuit 12, by the lighting interconnection 18, and this point of The reference is then transmitted by the interpretation circuit to the DCT calculation circuit in the form of a residual block, considered as a prediction error. The motion interconnection 13 simultaneously transmits the motion vector calculated for this processed macroblock and the corresponding mode type. In the decoder, the reconstructed block will be that corresponding to the motion vector to which the residual block is added. A third improvement of the device consists in adding an interconnection of regions 19, which makes it possible to transmit the reference point of the region to the quantizer 15.

This region interconnection 19 receives the reference points coming from the processing circuit 11, exchanges the information related to the coding of the current block with the image interpretation calculation circuit 12 and transmits the region information to the quantizer 15. This interconnection or interface divides the image into regions as a function of the modeling information that originates from the processing circuit 11. A label is assigned to each region or more exactly to each block, as a function of the region to which it belongs, the data regarding the block division of the image, which originates from the circuit 12 for calculating image interpretation. The quantization interval calculated by the quantization circuit 15 is modified as a function of this label, in such a way as to transmit a reference point of the region "carried" by the quantization interval to the decoder. This modification may depend on the characteristics of the regions: the reduction in the quantization interval for the non-uniform regions and / or the regions of low movement, increase the quantification interval for highly textured regions or regions of large movement. In this way, the quality of the decoded image is improved, for a given bit rate, allowing the fact that the defects in the blocks are less noticeable for the areas that are highly textured or of great movement. The same techniques as for motion vectors can be used for ambiguous membership cases: a block is declared to belong to the main region if the latter is for the most part in this block (majority mode), or even a block is declared to belong to the main object of the image if this object is part, even partially, of this block (main object mode). This interconnection makes it possible to improve the operation and the possible applications of the encoder: the subjective improvement for example by virtue of the reduction in the quantization interval for non-uniform zones or areas of little movement, as has been recently observed, two-dimensional applications or three-dimensional, the improvement in interactivity with respect to the decoder by virtue of the reference point of the region. Of course, these are illustrative modalities and the various alternatives can be combined or taken separately according to the desired image quality, the intended application, the processing speed or the desired degree of compression. In this way, the device of Figure 2 can very well construct functions of the device of Figure 1 and vice versa. For example, the circuit for the interpretation calculation and for the division into image blocks 12, can also carry out the functions of the subtractors 4i for the calculation of differential blocks or residual blocks based on the information coming from the circuit 13 of motion compensation, as well as the functions of the mode selector for the provision of circuit 14 DCT with the image block. The number of operating modes is thus increased, of course to the detriment of the complexity of the circuits. The processes and devices described are. fully compatible with MPEG-like data compression standards with respect to transmitted compressed data and decoding. An application in the present patent is the synthesis of the images, be it during production (virtual study, cartoons, video or film synthesis, special effects), for video games, for interactive applications or for virtual reality applications. The invention can be implemented in integral platforms (workstation, game console), given the level of simplicity obtained by virtue of the invention.

Claims (13)

1. A process for compressing digital data from a sequence of synthesis images that describe a scene that is the subject of a scenario, comprising a processing step to model the scene based on the mathematical data, a stage of image interpretation to create a synthesis image from this modeling, and a division of this synthesis image into image blocks, a differential coding of the current image block based on a block of at least one synthesis image, this block being defined based on in at least one motion vector, to provide a residual block, characterized in that the motion vector is calculated from the mathematical data emanating from the synthesis scenario, and that define the apparent movement of the various objects that constitute the scene, which is the object of the sequence.

2. The process according to claim 1, characterized in that the image interpretation is performed only for an image of N of the sequence, where N is a predetermined number, this image being coded in the intra mode and because the differential coding is performed on blocks of the intermediate images.

3. The process according to claim 2, characterized in that the current image block to be coded from an intermediate image is identical to the synthesis image block matched by the motion vector to provide a block with zero residue.

4. The process according to claims 1, 2 or 3, characterized in that the coding of the limit image blocks of an image is forced into the intra mode.

5. The process according to claim 1 or 2, characterized in that the illumination reference point for calculating the synthesis image is used to calculate a residual block for the inter-coding of the current block, as a function of the luminance disparity between the block of the previous image, matched by the motion vector associated with the current block, and the current block.

6. The process according to any of the preceding claims, characterized in that it performs a segmentation of the image, assigns a label to each of the blocks of the image, and modifies the quantization interval as a function of this label, to transmit a region reference point.

7. The process according to claim 6, characterized in that the quantization interval is increased for the blocks belonging to the regions that are highly textured or have great movement, and decreased for those that belong to the regions that are uniform or have little movement.

8. A device for the compression of digital data of a sequence of synthesis images, which describe a scene that is the object of a scenario, which comprises a processing circuit for the modeling of the scene, the images of which are to be synthesized based on the mathematical data, a circuit for the interpretation of the image and to divide the image into blocks, which receives the reference points from the processing circuit to make a synthesis image and divide the obtained image into image blocks, characterized in that it comprises a motion interconnection circuit using the mathematical data provided by the processing circuit and representing the displacement of the modeled objects constituting the scene, to calculate at least one motion vector associated with an image block, and which defines a predicted block based on which the coding of the image block is performed .

9. The device according to claim 8, characterized in that it comprises a circuit of compensation of movement of blocks of image, which receive the reference points coming from the processing circuit and the motion vectors coming from the circuit of interface or interconnection of movement, to provide predicted blocks, a subtractor to take the difference between the current block originating from the circuit for the interpretation of the image, and to divide it into image blocks, and the predicted block originating from the motion compensation circuit, to provide a residual block, a discrete cosine transformation circuit for the image blocks originating from the circuit, for the interpretation of the image and for dividing it into image blocks, or residual blocks originating from the subtracter, being the choice made by a mode selection circuit, as a function of energy criteria.

10. The device according to claim 8, characterized in that it transmits in the intra mode an image of between N images of the sequence, N being a predetermined number, this image being that which is the object of the calculation of interpretation by the circuit for the interpretation calculation, and to divide it into image blocks, and because other images or parts of images are transmitted in the inter mode by means of only the motion vectors that define the predicted blocks, and that originate from the motion interconnection circuit , the coding based on these blocks is corresponding to the blocks of zero value.

11. The device according to claim 8, characterized in that it transmits in the intra mode an image of between N images of the sequence, N being a predetermined number, this image being that which is the object of the calculation of interpretation by the circuit for the calculation of interpretation, and to divide it into image blocks, and because other images are transmitted in inter mode by means of residual blocks that represent the difference between a current block and a predicted block, and which are obtained from an interconnection circuit of lighting whose function is to calculate the difference in illumination between the current block and the block predicted based on the mathematical data.

12. The device according to claim 10 or claim 11, characterized in that it comprises an interconnection circuit or error interface, to detect the error between the current block and the predicted block, to force the coding of the current blocks to the intra mode when the error with the predicted block exceeds a certain threshold.

13. The device according to any of claims 9 to 12, characterized in that the digital data is compressed according to the MPEG-2 format or a derived format.