FR2930107A1 - Video sequence data processing method for e.g. video telemonitoring field, involves generating pattern image joining image patterns that simultaneously satisfy criteria, and storing generated pattern image in reference storage zone - Google Patents

Abstract

The method involves determining image patterns obtained from a set of image data, and selecting image patterns satisfying a predetermined criterion, among the determined image patterns. The determined image patterns that satisfy another predetermined criterion related to a reference storage zone i.e. buffer, are identified. A pattern image joining image patterns that simultaneously satisfy the criteria, is generated. The generated pattern image is stored in the reference storage zone. Independent claims are also included for the following: (1) a video sequence data processing device comprising a video coding unit (2) an information storage medium comprising instructions for performing a method for processing data of a video sequence (3) a computer program product comprising instructions for performing a method for processing data of a video sequence.

Description

The present invention relates to a method of processing data of a video sequence to be encoded and a method of processing data of a stream comprising an encoded video sequence, as well as the associated processing devices. It is known to transmit video sequences over a communication network between two remote communication apparatuses such as a client apparatus and a server apparatus in a client-server architecture. The contexts of video transmission between a client and a server are numerous. For example, videoconferencing and video surveillance. In many cases, the networks over which video data travels have limited resources, especially in terms of bandwidth. These constraints of limited flow impose to compress the video sequence before transmitting it. The video compression methods that are today considered most effective are based on motion compensation. Thus, so-called reference images of a video sequence to be compressed are used to compress other images of this sequence by performing a motion prediction from one or more reference images. In some applications, the memory size of the decoders capable of receiving and decoding an encoded video sequence is limited to a given number of reference images. These images are stored temporarily during a specified time interval and are deleted from the memory so that 30 new ones can be stored. This limited number of reference images does not allow optimal use of motion prediction. As a result, several image areas of the same video sequence are coded and transmitted repeatedly, which is not effective.

It follows, moreover, an increase of flow which can be problematic considering the limited resources in flow (data loss,

Moreover, in a video transmission system between a client and a server such as a videoconferencing or video surveillance system, the filmed scene includes fixed objects that may be partially concealed by one or more moving objects within the scene. Thus, during a video conference, a person can leave the field of the camera and then return later. Generally speaking, during the same video sequence, moving objects can appear and disappear several times. However, as indicated previously, the number of reference images stored temporarily is limited. As a result, the pixels constituting the images of the objects moving out of the field of the camera are removed from the memory after a determined time interval.

Thus, when these same objects return later in the field of the camera, there is no longer in the memory of reference image (s) useful (s) to facilitate the coding of images of these objects. The coding of these images must therefore take place in intra mode, which results in an increase in the rate.

Document US 6,678,417 discloses a method of compressing and transmitting a video sequence from an image dictionary that is created dynamically from the images of the sequence. Each image of the dictionary is associated with an image index. Thus, each new image of the sequence is compared with the images of the dictionary and, if the new image corresponds to an image of the dictionary, then only the image index is transmitted. If part of the new image corresponds to an image of the dictionary, the difference between the two images is calculated and then transmitted with the corresponding image index.

This method makes it possible to reduce the bit rate necessary for the transmission of a video sequence by using a scalable dictionary shared between a server and a client.

However, this method does not solve the problem of the limited memory size for the storage of reference images in the context of video coding based on a motion prediction. The methods just described are therefore not entirely satisfactory. On the other hand, more generally, it would be interesting to have a new method, and a new associated data processing device, which more effectively manages the compression of the images of a video sequence by prediction of movement. The present invention aims to remedy at least one of the aforementioned drawbacks and / or to achieve one of the aforementioned objects by proposing a method for processing data of a video sequence comprising a succession of images, characterized in that the method comprises the following steps: - video coding of images of the sequence based on a motion prediction from reference images of the sequence, reference images being stored in a so-called reference storage area, - determination of patterns each obtained from at least one set of data constituting an area of an image, - selecting from the previously determined image patterns, image patterns satisfying a first predetermined criterion, - identifying the previously determined image patterns that satisfy a second predetermined criterion related to the reference storage area, - generating at least one pattern image uniting image patterns that satisfy both the first and second predetermined criteria; storage in the reference storage area of said at least one generated pattern image. Correlatively, the subject of the invention is a device for processing data of a video sequence comprising a succession of images, characterized in that the device comprises: means for video coding images of the sequence based on a motion prediction from reference images of the sequence, reference images being stored in a so-called reference storage area, - means for determining image patterns each obtained from at least one set of data constituting an area of an image, - means for selecting, from the previously determined patterns, image patterns satisfying a first predetermined criterion, - means for identifying previously determined image patterns that satisfy a second predetermined criterion related to the reference storage area, - means for generating at least one image of patterns comprising image patterns that satisfy both the first criterion and the second criterion 10 predetermined, - storage means in the reference storage area of said at least one pattern image generated. By generating one or more images that group together certain patterns (those meeting the two predetermined criteria) and storing them in the reference storage area, the limited storage resources are more efficiently managed. Indeed, previously, such patterns were often stored in isolation in the reference area in the form of several reference images which thus occupied a relatively large storage space. Thanks to the invention, these patterns are gathered in at least one pattern image which occupies a reduced size memory space. Moreover, this generation of pattern images is performed dynamically over time, so that patterns selected at a given moment to be part of a pattern image, because satisfying the criterion (s) 25 predetermined, may not be selected later. The identification of the patterns conforming to a second predetermined criterion is obtained by tracking the evolution of the patterns with respect to their current or future presence in the reference storage area. In particular, it is determined whether or not the patterns are already present in the reference area and, if present, whether they are about to be deleted from the area to allow storage of new data. reference image. It should be noted that the selection and identification steps can be reversed.

According to one characteristic, the determination of image patterns is performed from encoded images. Because the patterns have been identified during the coding process, data processing is faster.

Furthermore, the image encoding operation of the sequence itself constitutes a preselection of patterns since it will thus determine those which have already served as reference image data (sets of image data used as references for motion prediction when video coding images). As a result, these patterns determined during coding will have a greater chance of satisfying the first predetermined criterion. As a first variant, the patterns can be determined independently of the motion prediction, for example, from a region analysis of an image. In addition, according to a second variant, some of the image patterns can be determined independently of the motion prediction and other image patterns from encoded images of the sequence. According to one characteristic, the first predetermined criterion reflects the importance of an image pattern for predicting image motion (s) of the sequence. The importance of a pattern has an impact on the efficiency of the motion prediction coding between the images. Thus, by keeping the important patterns in the reference storage area, the prediction of the images of the sequence is improved. In addition, when the encoded images are to be transmitted, the amount of data to be transmitted is reduced. According to one characteristic, a pattern responds to the first predetermined criterion if it is used several times to predict the motion of images of the sequence. This criterion defines a particular case of significance of a pattern. Other particular criteria of importance can be envisaged. According to one characteristic, the second predetermined criterion reflects the imminence of an image pattern to be deleted from the reference storage area or its absence from this area. The identification of image patterns thus leads to identifying: - the patterns that are already present in the reference area and which will be deleted (for example because of the automatic storage management of the memory space) in order to storing new reference image data and - the patterns that are not yet present in the reference area (for example, patterns that have been generated independently of the prediction coding or patterns determined from the first images of the sequence). Note that we can refine the identification of the patterns that are about to disappear from the reference storage area by determining in time the first patterns that are expected to disappear, then the second and so on (management forward). According to one characteristic, the number N of image patterns combined in the said at least one pattern image is determined as a function of the memory capacity of the reference storage area that will be used during the decoding of the coded video sequence.

Thus, the number N is even larger than the size of the reference storage area is greater. According to one characteristic, all the image patterns relating to the same image of the sequence are combined in the same pattern image. This arrangement makes it possible to optimize the storage management of the patterns and also to facilitate the decoding operations. According to one characteristic, the method comprises a step of positioning an image pattern in said at least one pattern image so as to minimize the coding cost of the motion vectors. According to one characteristic, the method comprises a step of generating marking information distinguishing said at least one pattern image generated from the images of the sequence. This marking will be useful during decoding operations to identify images that should not be displayed. According to one characteristic, the method comprises a step of encoding said at least one generated pattern image. The encoding of the generated pattern image or generated pattern images makes it possible to process this image or these images at the decoder as any image of the video sequence. It is thus possible to use a conventional decoder.

According to one characteristic, the method comprises a step of transmitting on a communication network of said at least one generated pattern image. According to one characteristic, the method comprises a step of transmitting the encoded video sequence with said at least one generated pattern image. According to one characteristic, the method comprises a step of long-term marking of a pattern image indicating that this image must be maintained in the reference storage area as long as the marking is present. The tagged image is thus automatically retained, which facilitates management of the reference storage area. The invention also relates to a method for processing image data of a stream comprising an encoded video sequence comprising a succession of coded images according to a coding based on a motion prediction based on reference images of the video sequence, characterized in that the method comprises the following steps: - reception of the image data of the stream, the image data comprising coded images and at least one image, coded or not, consisting of image patterns , each image pattern having been obtained from a set of data constituting an area of an image, said at least one image pattern image satisfying both a first predetermined criterion and a second predetermined criterion related to the reference storage area; decoding of coded images, in particular from said at least one image of patterns for display, said at least one image of patterns not being as displayed. Correlatively, the subject of the invention is a device for processing image data of a stream comprising an encoded video sequence comprising a succession of coded images according to a coding based on a motion prediction based on reference images of the video sequence, characterized in that the device comprises: means for receiving the image data of the stream, the image data comprising coded images and at least one image, coded or not, consisting of patterns of images, each image pattern having been obtained from a set of data constituting an area of an image, said at least one image pattern image satisfying both a first predetermined criterion and a second predetermined criterion linked to the reference storage area; coded image decoding means in particular from said at least one pattern image for display, said at least one pattern image; being not displayed. Thus, at the decoder, the pattern image (s) are used for the decoding of the images of the video sequence but are not displayed. The presence of this pattern image or these pattern images in the associated reference storage area of the decoder makes it possible to efficiently and optimally use the limited size available memory space. According to one characteristic, the received stream also comprises marking information distinguishing said at least one image of image patterns from the images of the sequence. This information is useful for the decoder to identify the images of the stream that should not be displayed. According to one characteristic, the method comprises a step of selecting, from the received stream, images that can be displayed, the selection being made according to the marking information. Thus, a selection step is set up at the decoder from the tag information present in the received data stream. Other characteristics and advantages will become apparent from the following description, given solely by way of nonlimiting example and with reference to the appended drawings, in which: FIG. 1a schematically illustrates the principle of video coding; based on a motion prediction; - Figure 1b schematically illustrates an example of communication devices connected by a communication network; FIG. 2 is a general schematic view of an algorithm detailing the main operations performed by a data coding and transmission apparatus; FIG. 3 is a general schematic view of an algorithm detailing the main operations performed by an encoded data reception and decoding apparatus transmitted by the aforementioned coding apparatus; FIG. 4 is an algorithm detailing the main steps performed during the analysis operation 202 of FIG. 2; FIG. 5 is a general schematic view of an algorithm detailing the main steps performed during operations 204 and 206 of FIG. 2; FIG. 6 is a general schematic view of an algorithm detailing the main steps performed during encoding 208 of a pattern image of the algorithm of FIG. 2; FIG. 7 is a general schematic view of a device in which the invention is implemented.

The invention relates to video data processing in which the data of the different images constituting the video are predictively coded by motion estimation from some of the images of the video. The principle of such encoding is illustrated in Figure la on which a video sequence referenced 100 comprises a set of images 101, 102, 103, 104 ... which will be encoded by a coding device also called encoder. Different compression standards such as the H.264 and MPEG4 part 2 standards use an encoder implementing a motion compensated encoding.

The coding of the images of the sequence 100 is carried out according to several modes of coding. Indeed, the first image 101 can not be encoded using a reference image and is therefore encoded without motion compensation. This is called intra or I.I coding.

On the other hand, the image 102 is coded according to an encoding mode called inter or P which is based on a motion compensation performed from a reference image. The coding P is made from a single reference image, this image being a past image. Thus, the image 102 is encoded from the past image 101.

The image 103 of the succession of images is, what is called, a bi-predicted image or B, that is to say that for its encoding and thus to be able to perform the motion prediction, the encoder must have images 101 and 102. An image B is predicted from two past images or two future images or from a future image and a past image.

Thus, as explained above, the first image 101 of the sequence is first encoded in intra mode, then the following successive images 103, 104 and 102 are stored before being able to encode the image 103. It will be noted that the images 101 and 102 are stored in decoded form after being already encoded. The encoding order of the images is therefore 101, 102 and finally 103. The reference images 101 and 102 are stored to enable the coding of the image 103. It will be noted that, in the example which has just been described , image coding based on motion prediction is performed from past (101) and future (102) images. In this case, it should be noted that the encoder must store a number of reference images before encoding certain images. In this respect, it may be envisaged to encode each image on the fly by using, as reference images, only the previous images, namely those which appear in the video sequence at previous times. In such a case, the video coding scheme will be of the IPBBB type. This scheme is suitable for on-the-fly encoding if and only if the B-images are predicted from past references. It should be noted that this type of encoding is the one most frequently used in videoconferencing systems. It makes it possible to reduce the delay between the time of capture of the image, on the server side, and the moment when the received and decoded data are displayed, on the client side. Note that the images are processed by elementary unit of image data which are, for example, macroblocks (blocks of 16x16 pixels). As represented in FIG. 1a, the image 103 is divided into several slices 0, 1, n (known in English terminology as 30 slices) and each slice has the structure of the slice 108. This slice groups together several elementary units of data (macroblocks) MBO, MB1 MBn.

A storage area 110, referred to as a reference area, constitutes a memory space that is present on both the client device and the server apparatus side. This storage area, for example, a temporary one (called buffer in English terminology) makes it possible to store a set of images of the video sequence 100 which are used as references during the coding by motion compensation. The images present in this memory zone and denoted 1, 2, 3, 4, in the example are decoded. More particularly, the images of this memory zone are images that have been encoded and then decoded, both at the encoder and at the decoder. Generally, this memory zone is filled according to the principle of filling of a memory FIFO type (acronym meaning in Anglo-Saxon terminology First ln First Out) according to which the new reference images (image 4) are added in the reference memory zone in queue. When the memory area is full, the image at the top of the queue (image 1) is removed from the area. This management of the reference image storage is detrimental to the coding of the images of the sequence. Indeed, subsequent images of the sequence to be encoded will no longer be able to use certain reference images that have been deleted from the area 110. The subsequent images will therefore have to be coded in intra mode, thus requiring an increase in throughput. The temporary reference storage area 110 is used in the encoder for estimating or predicting the motion of certain images from reference images or reference image data present in this memory area. In the decoder, this memory area is reconstructed as the coded image data is transmitted and is used to perform motion compensation from the reference images to obtain compensated images.

As represented in FIG. 1b, the exemplary embodiment that will be described is part of a client-server architecture. A server-type communication device 150 receives a video sequence from a video data acquisition apparatus 152, encodes it and transmits it over a communication network 154.

The encoded video sequence is received by a client type communication device 156 which then decodes it for display. The communication device 150, as well as the device 156, has a structure of the type of that represented in FIG. 7 which will be described later and in which the various means necessary for the implementation of the invention, both on the server side only on the client side are integrated. The device 150 stores, for example, a computer program which, during its execution, makes it possible to implement the method according to the invention on the encoder side. This computer program is based on an algorithm schematically shown in FIG. 2. Similarly, the device 156 also includes a computer program which, when executed, enables the method to be implemented according to FIG. the invention on the side of the decoder. This program is based on an algorithm shown schematically in Figure 3 and will be described later. It will be noted that the data once coded by the device 150 may, alternatively, be stored before their transmission. Similarly, storage may alternatively be considered upon receipt of encoded data transmitted over the network 154 before being decoded, or, after decoding before being displayed. As a variant, the data once processed according to the invention, on the server device side, can be stored on a storage medium such as a disk for possible subsequent transmission. The storage of processed data according to the invention does not require a large memory size. As shown in FIG. 2, during a first step 200, the apparatus 152 of FIG. 1b captures images constituting a video sequence, for example, during a video surveillance or videoconference type application. , and then transmits them to the device 150. More particularly, these video data are transmitted to the encoder of the device 150, during a step 202, in order to be encoded.

The encoding of the images of the sequence takes place as briefly indicated in the description of the figure la, image by image, from one or more reference images, except for the first. The use of images I is not limited to the first image. During the sequence, some images are encoded in intra coding mode when the motion prediction has a higher coding cost than the intra encoding. Note that there is no limitation for the choice of the reference image for motion prediction: the use of past images, future images or a combination of one or more past images and one or more future images is possible. When video coding the images of the sequence, reference image data is stored temporarily in a reference storage area such as area 110 of Fig. La. These reference image data are reference images which are real images of the video sequence and also other types of images (pattern images) which will be described below and which are not images of the image. video sequence. This reference storage area is of relatively limited capacity and is present in the apparatus 150 as well as in the apparatus 156 of FIG. 1b.

The images encoded in step 202 are then processed, for example, successively one by one during the following steps 204, 206 and 208 rather than by series of images. This makes it possible to prevent image patterns from being deleted from the reference storage area before having been able to process the last image (s) of a series of images. In fact, in such a case, the image patterns deleted from the storage area will have to be coded in intra mode, which will lead to a higher coding cost. Alternatively, a pattern image may be generated at a regular time interval, for example, after a given series of images of the video sequence. In this case, a series of images of the video sequence are encoded before being analyzed (step 204) and one or more pattern images are then generated. The advantage of this variant lies in the fact that the selected patterns (step 206) are the patterns that have been judged most important over a longer period of time. During the analysis step 204, one or more image patterns are determined from an encoded image.

Each image pattern comprises a set of data (e.g., pixels) that constitutes or represents an area or region of the image. The pattern may alternatively be a set of image data not present in the image being processed or analyzed but obtained from a set of data corresponding to a region or region thereof. image, or even several of these sets. The determination of this pattern is then performed independently of the motion prediction. The pattern in question will be, for example, a set of image data close to several portions of the same image or several portions of different images, and the data of this pattern may then be an average of the data from the different images. image portions. It should be noted that the image pattern (s) determined during this step are determined because they have the particularity of being used as references for the motion prediction during the video encoding of the image (s) of the sequence . It will be noted that during this step the motion vectors of the current image are extracted in order to identify the areas of the reference images that are used as references during motion compensation. Each of these areas is a motif. For type B images that have two references, two new patterns are extracted by predicted macroblock. In summary, during this analysis step, one or more image patterns are determined with respect to the image being processed: either by extracting this or these patterns from the image under consideration, or by associating a plurality of image patterns or a plurality of images having an overlapping area, the pattern resulting from the association then replacing the associated patterns, or by synthesizing the one or more patterns from sets or portions of data of this single image or of this image and other previous images. For example, the generation of a pattern can be achieved by averaging the pixels of several existing patterns.

During this step, the image patterns previously determined and which satisfy a predetermined criterion are also selected. More particularly, a value is determined which will be associated with each previously determined image pattern.

This value is obtained for each image pattern according to this predetermined criterion. This criterion translates, for example, the importance of a pattern for the prediction of image motion (s) of the video sequence. The value thus determined is called the value of importance of the reason 10 considered. For example, a pattern will be associated with an importance value when this pattern is used more than once as a set of reference image data for the prediction of image motion (s). It will also be assigned an importance value when this pattern is part of an image that disappears from a scene of the video sequence at a given moment and reappears later during the course of the sequence. Thus, the step 204 receives, as input, an encoded image of the video sequence and provides, at the output, one or more determined patterns related to this video image, as well as a value (for example an importance value). translating the behavior of the pattern with respect to a predetermined criterion. The one or more image patterns are represented by a structure containing the data of the one or more patterns and the aforementioned value is generally an integer. It will be noted that, at the end of step 204, the one or more patterns that have just been determined with respect to the image being analyzed and their associated value are available. Patterns determined from the previously processed images of the video sequence and their associated updated values are also available. We will come back later in more detail, in the description of Figure 4, this step 204 image analysis. The algorithm comprises a next step 206 of determining the image patterns determined according to another predetermined criterion. During this step, it will more particularly proceed to the identification of the image patterns whose storage in the temporary storage area 110 reference of the device 150 must be maintained.

To do this, it is first necessary to identify, in the temporary reference memory space 110, the previously determined image pattern (s) that are present in this memory space. Then, among the pattern or patterns identified as being present in the memory space 110 (for example, the patterns having the largest values of importance), it is intended to select one or more image patterns according to this other criterion. predetermined. More particularly, this selection criterion reflects the imminence for a given image pattern to be deleted from the reference memory space 110.

Indeed, the images present in this memory space are regularly deleted at the expiration of a predetermined time interval. Note that during this step we also identify the image patterns that are not yet present in the area 110 but should be stored there.

The pattern or patterns determined in step 206 and that one wishes to keep as reference image data in the temporary storage area 110 or that one wishes to store in this area will be inserted in a virtual or artificial image. In this way, from the patterns determined in step 206, one or more pattern images are generated and then stored in the area 110. Each pattern image combines one or more patterns of patterns. images determined with respect to the same image or with respect to several different images of the sequence. It should be noted that the image or images thus generated have no reality in the video sequence in the sense that they do not represent any instantaneous image of this sequence (image carrying information representative of the semantic content of the video). In other words, the pattern images can be considered as computer-generated images obtained from images or parts of real images of the video sequence.

Note that the pattern image or images generated in step 208 (several pattern images are generated when the area occupied by all the patterns selected in step 206 is larger than the chosen size of a pattern image ) will be used to maintain the image patterns in the temporary reference storage area 110.

Steps 206 and 208 will be described in more detail when describing the algorithm of FIG. 5. The image or pattern images are then encoded in step 202 by the device encoder 150. and transmitted over the network to the client apparatus 156 in step 210. Note that the pattern image or pattern images are not necessarily encoded before being transmitted. However, pattern images must be encoded if you do not want to modify a conventional decoder. Indeed, in this case, the pattern images generated in step 208 will be considered images of the stream and conventional decoding will be applied. On the other hand, if a new codec has a set of commands for storing zones of several images in a new image of the reference storage area, it is not necessary to transmit the pattern image and thus to encode it. . It will be noted that the image or the images of transmitted patterns are accompanied by marking information which makes it possible, when receiving these image data, to differentiate the image or images of patterns (artificial image (s)). )) real images of the video sequence. In the exemplary embodiment described, each image of the sequence is encoded, where technically possible, by motion prediction from one or more previously encoded images used as reference image data. The image or the images of patterns must be interposed within the images constituting the compressed video sequence and which is transmitted to the device 156. Indeed, the presence of images of patterns in the stream transmitted to the device 156 allows the latter to use, as and when, the received pattern image or images in order to reconstitute the useful reference image data in its reference temporary storage area identical to that of the device 150. It will be appreciated that the pattern images and those of the video sequence can be sent separately or not. Pattern images are fundamental for decoding since they are reference images that will be used to decode subsequent frames of the video clip. The advantage of transmitting the images separately is to be able to apply an error correction method (for example of FEC type meaning in English terminology forward error correction) on the images of patterns in order to limit the transmission errors on these key images for the decoding of the stream.

It should be noted that the images of the sequence which have just been encoded in step 202 can be transmitted immediately to the client during step 210, in parallel with their transmission to the processing unit which implements the Steps 204, 206 and 208. Thus, the sending of the data is better distributed over time, which makes it possible to avoid problems of network congestion. However, in some applications it may be preferable to transmit the pattern images with the images of the sequence. Alternatively, the encoder of the device 150 may perform motion estimation with respect to future images as shown in Fig. 1a for the image 103. In this regard, the pattern images need to be integrated with the image. good position in the video sequence so that the receiver device 156 can decode the thus predicted images without loss of time. To do this, the encoder must save a certain number of images 20 before proceeding to their encoding as already explained with reference to FIG. When an image refers to future images, it is sufficient to force the encoding moment of a pattern image before the images of the video sequence. Thus, sending a pattern image at the end of step 208 to encoder 202 is specifically processed: each pattern image received by the encoder is encoded immediately. For example, for the encoding of the sequence described in FIG. 1a, the encoding order of the images is 101, 102, 103, 104. Once the image 101 has been encoded, the images 103 and 104 are stored in an image. temporary memory to be encoded after receiving image 102. When a pattern image 30 is generated in step 208 after encoding image 102, this pattern image is encoded before images 103 and 104 As indicated above, the algorithm of FIG. 3 is implemented in the device 156 after receiving a data stream therefrom. This stream comprises an encoded video sequence transmitted by the device 150, as well as one or more images of image patterns, coded or not, transmitted separately from the stream or not and the associated marking information. The tagging information is transmitted either by adding a parameter in the header slices (slices in English terminology) of the image or by using a separate message that indicates the number of the image not to display. This message is defined using the RTP protocol (known in English terminology as the Real Time Protocol): an RTP packet whose payload in English terminology contains the number of the marked image is created and sent to the client. The client then uses this RTP packet to determine the marked images. This latter method is used in the preferred embodiment of the invention. If the pattern images are sent separately from that of the video sequence, then the marking is implicit. Thus, the image data that is encoded is, for example, transmitted in the form of data packets and is received in step 300 of FIG. 3 by the device 156. In the next step 302, the images Encoded encodings are decoded and, when the pattern images are also encoded, the same is done with their decoding. It will be noted that the images of the video sequence and the images of patterns are thus treated in the same way during their reception and their decoding. In the next step 304 data processing is performed using the received tagging information. This marking information makes it possible to select from the received data stream between the real images of the sequence and the marked pattern images which, they do not have to be transmitted to the display during the display. Step 306. Thus, if the decoded picture is a reference picture it is stored in the reference storage area of the client device 156 which is identical to the area 110 of Fig. la.

Since the storage memory of the client device has a limited size, the management principle of the occupation of a FIFO type memory is applied when the reference storage area is occupied to the maximum of its capacities.

In the next step 306 the actual images of the video sequence are thus displayed. The algorithm of FIG. 4 corresponds, for example, to a sub-program which details the operations carried out during step 204 of FIG. 2. As explained previously with reference to this figure, each encoded image of the video sequence is treated successively. For each new image analyzed in step 204, one or more new image patterns are determined from the current image (being processed) and possibly from images previously appearing in the sequence and assigned to them. importance values. Thus, at each iteration, the set of image patterns obtained for the previous images and the importance values associated with them are incremented. It will furthermore be noted that the importance values associated with the previously obtained image patterns are updated taking into account the new image patterns determined for the current image. It should be noted that, when executing the algorithm of FIG. 4, it is intended to determine the state of the reference storage area 110 of FIG. 1 to 20 and to determine which images of the sequence are used. as references. The reference storage area is filled during the encoding of the images in step 202 of FIG. 2. In order to be able to use this area during step 204 of FIG. 2 (thus during the execution of the algorithm of Figure 4), two methods are considered. According to a first method, it is envisaged, during step 204, to decode the images encoded in step 202 and sent by the coder to the analysis unit and, thus, to reconstitute the storage area of reference 110 within the analysis unit. This step of decoding and updating the reference area is carried out for all the encoded images, including the previously generated pattern images, and constitutes the first step 400 of the algorithm of FIG. first coded picture of the sequence is decoded and, if it constitutes a reference picture, then it is stored in the area 110.

It will be noted that it is possible to proceed image by image or else image packet by image packet, as already described above. According to a second method, the coder of the device 150 sends to the analysis unit of the image performing the operation of step 204, not only each image encoded in step 202, but also the state of the image. reference temporary storage area 110 when encoding this image. Note that for the H.264 standard the long-term memory area and the short-term memory area are both considered to be in the reference storage area. However, there are other codecs, including an extension of the standard H.264 (SVC for Scalable video coding) that can use several reference storage areas. In this case, the invention can be used in application to each reference zone. The second method has the advantage of not requiring additional calculations, while the first method makes it possible not to modify the operation of the encoder, which is particularly advantageous for a hardware type encoder, ie -This is not software. If a second method is retained, then the algorithm of the figure does not execute step 400 but proceeds directly to the next step 402. During this step, new patterns are determined: by extraction of an area of an image, - by association of several motifs having overlapping between them, or - by creation of a new motif. Note that the patterns considered in the invention may correspond to areas or portions of images used as references when motion prediction encoding other images. However, the patterns are not necessarily areas used as references. Indeed, it is possible to extract an area of an image for example by segmentation and to consider it as a pattern, independently of the motion prediction coding. As already explained above, a pattern may be a set of data constituting an area of an image or be obtained from such a set or from several sets of data constituting several areas of the same image or several different image areas.

When several sets of image data, constituting several areas (several sets of pixels) of the image, are identified, these patterns are, for example, associated to form a single larger size pattern, rather than keeping multiple patterns that overlap with each other. For example, new pixels adjacent to that pattern can be selected for a given pattern that is identified in an image to obtain a pattern of a size multiple of 16x16 pixels. Thus, in the example considered, the patterns taken into account have a shape, for example rectangular, and a size that is multiple to that of a macro-block (16 × 16). Alternatively, several patterns may be replaced by a single pattern whose pixel value is the average of the pixels of each of the plurality of patterns. The pattern obtained by this grouping of the plurality of patterns will then consist of synthesized pixels. An image pattern in the sense of the invention is for example represented by a data structure composed of the following elements: a dynamically created array of pixels containing the pixel values of the pattern, an integer indicating the width of the pattern in number of macroblocks (unit allowing to define one of the dimensions of the pattern), - an integer indicating the length of the pattern in number of macroblocks (unit making it possible to define this other dimension of the pattern), - a date indicating the moment when the pattern was used for the prediction of image motion (s) for the last time. It should be noted that, when analyzing an image, it is sufficient to extract existing areas from the image using an image region analysis tool or a segmentation tool. Each region extracted by the tool is an image pattern in the sense of the invention. It should be noted, moreover, that the patterns can adopt complex shapes and various sizes.

The patterns that are determined in step 402 are systematically stored in a dedicated storage space of the apparatus 150, independent of the reference storage area, as long as the memory space is not full. The algorithm of FIG. 4 comprises a step 404 which makes it possible to select, for the current image, an image pattern from those previously determined. This selected pattern will then be processed during steps 406, 408 and 410 and a processing loop will thus be performed to take into account all the patterns previously determined from the beginning of the video sequence.

In step 406 a test is performed on the selected pattern to determine if this pattern is likely to be used as reference image data for the motion prediction of the image being analyzed. This test can be answered positively in two cases. In the first case, we find all the image data (pixels) of the selected pattern in the image being processed. In this case, the importance value associated with the pattern is incremented by one unit in the next step 408 and the date of this pattern is updated (date indicating the moment when the pattern was last used). Note that the importance value of a new pattern generated from a set of patterns will be equal to the sum of the importance values of each of these patterns. The other case is where the pixels of the considered pattern do not correspond to any area of the reference image, that is to say that this pattern is not found in the identical image. In this case, if the pattern is close to a reference area used to predict a macroblock of the current picture, then this pattern is considered to be the best candidate for the motion prediction of a picture area. the current image. To do this, the motion vectors are extracted from the current image. These motion vectors make it possible to identify the area used as a reference in the reference image. Then, the reference area is compared to the selected pattern. This reference area and this pattern are considered close when the sum of the absolute differences of their pixels is less than a threshold determined by the user.

In this second case, when the pixels of the selected pattern and those of the reference area are close, then the importance value associated with the pattern being processed in step 408 is also incremented and the date of the pattern is updated. .

The next step 410 is a test performed on the set of previously determined patterns. This test plans to go to step 404 already described in order to select the following patterns in this set. Note that the patterns are, for example, classified according to the order in which they were determined during the processing of successive images of the sequence. Back to step 406, when it is found that the pattern being processed is not used as a reference for the motion prediction of the current image, i.e. has nothing in common with this image, so we go directly to step 410 which has just been described. It will be appreciated that patterns may be eliminated from all the patterns determined in step 204 of FIG. 2 and stored in a dedicated space of the apparatus 150. Thus, for example when the As the maximum number of patterns is reached relative to the size of the corresponding memory space, it may be decided to delete a pattern in order to add a new pattern. In this regard, one may choose to eliminate the least important reason (s), ie, those which have not been used for the prediction of movement during a specified period (long period). . Thus, the oldest pattern (s) having the lowest importance values with respect to the other given patterns are deleted. It can be seen that the processing of each image of the video sequence contributes to identifying the motifs of the video sequence which are the most important among all the determined patterns. Indeed, during the processing provided in the algorithm of FIG. 4, if a pattern is present in several images of the video sequence, its importance value will be incremented several times during step 408. Moreover, if a pattern corresponds to an object of the scene of the video sequence that leaves this scene and returns there after a certain time, then this motif will be given an increasing importance to each of its appearances. This will justify its maintenance in the reference storage area 110 of FIG.

It should be noted that the invention makes it possible to qualify the importance of a pattern with respect to a set of patterns. At the end of the execution of the algorithm of FIG. 4, a set of image patterns and their associated importance values which have been updated during the processing of the current image are thus obtained.

The algorithm of FIG. 5, which will now be described, details the successive operations of steps 206 and 208 of FIG. 2 applied to the image pattern (s) determined in FIG. 4. The context of FIG. DESCRIPTION OF FIG. 5 is that of the processing of a current image of the video sequence and considering that an image of image patterns has already been generated by executing the algorithm of FIG. 2 for the preceding images. In FIG. 5, a reference storage area 500 is represented in a state corresponding to that of the processing of the current image. It can be seen that this zone 500 comprises a pattern image 500a combining several image patterns and two reference images 500b and 500c. These three images constitute reference image data which are used as references for the prediction of images of the video sequence. In this figure is also represented the set 520 image patterns previously determined and stored in a reserved memory space.

The reference storage area 500 is used by the encoder, as well as by the decoder (which will have reconstituted on its side), while the set of patterns 520 is only available to the encoder. In the example considered, the reference zone 500 has a maximum size of three images.

The pattern image 500a has been previously created, the reference image 500c has been previously decoded and the reference image 500b is the oldest stored reference image. This last image is destined to disappear insofar as the short-term storage area 500 is managed as a FIFO-type memory.

Note that for the H.264 compression standard, the pattern image 500a will be replaced by a new pattern image generated. Thus, the images 500a and 500b will be deleted from the reference area 500, while the image 500c will be retained.

Regarding the MPEG4 part 2 standard, the principle of FIFO memory management is used for the buffer temporary storage area and only the 500b image will be deleted. The purpose of the algorithm of FIG. 5 is to determine, among the patterns of the set 520, those which must continue to be stored in the reference zone 500 and to generate one or more images from the patterns thus selected. The algorithm of FIG. 5 comprises a first step 502 providing for recovering, among the set of patterns 520, a pattern (for example in the chronological order of obtaining these patterns) in order to determine, during the following steps, if it is to be selected to be held in the storage area 500. In the next step 504, a test is performed to determine if the pattern being processed satisfies a predetermined criterion. It should be noted that a value has already been assigned to each pattern as a function of a predetermined criterion when executing the algorithm of FIG. 4, for example, importance. The importance is, for example, as has been seen previously the number of times a pattern is used to perform an image motion prediction. An importance factor can also be retained if a pattern reappears several times over time in the sequence. On the other hand, a pattern that is only very rarely, if ever, used for image motion prediction, is considered to be of little or no importance. Here, the test consists in selecting more particularly the patterns which are the most important among the set of previously predetermined patterns 520. For this purpose, a number N of image patterns considered to be the most important for the motion picture prediction of the sequence is selected.

The value N is, for example, estimated according to the memory capacity of the decoder. Indeed, the larger the maximum size of the reference storage area allowed by the client device 156, the greater the value of N will be.

Another method for determining the most important reasons may be to calculate a threshold value of importance according to the capabilities of the client device, for example, depending on the memory capacity of the decoder. According to this second method, the pattern will be considered important if its significance value exceeds the aforementioned threshold value.

When the pattern being processed (current pattern) is selected according to this predetermined criterion, then the next step 506 is taken. In this step, a test is performed to determine whether the current pattern is present in the current pattern. an image stored in area 500 (image 500a and image 500c for MPEG4 part 2, for the H.264 standard only image 500c is retained.). Indeed, during this step it is taken into account that the image 500b will be deleted, given the normal management of operation of the memory zone 500. Thus, if the pattern is not present in the one 500a or 500c images that will normally be preserved taking into account the operation of the memory zone 500, then this important reason must be marked in the next step 508. Two cases correspond to this possibility: the case where the important reason is in the image 500b which is about to be deleted from the reference area 500 and the other case where the pattern is not present in the reference area 500.

The algorithm of Figure 5 thus aims to determine, among the most important reasons (first criterion), those who are expected to disappear from the reference area and who are not there yet (second criterion). It should be noted that the other cases where the pattern being treated is not marked correspond to the cases where: the reason is not considered important (negative answer at the end of the test at step 504) and the pattern being processed is already in one of the images of the reference zone 500 which is not about to be deleted (positive response after the test of step 506). At the completion of steps 504, 506, and 508, step 510 is performed to determine whether all the patterns of the set 520 have been processed. If not, step 502 is again described to recover the next pattern of this set and apply step 504 and optionally steps 506 and 508. If so, the determination step (step 206) of image patterns according to a predetermined criterion is completed. All the selected motifs have been marked. Note that if at the end of step 510 no pattern of set 520 is marked for the image being processed, then no pattern image is created for that image. The algorithm of FIG. 2 then returns to step 202 to process the next image. Note that, in the context of the H.264 compression standard, if the 15 patterns that have just been selected at the end of steps 502 to 510 are already included in a pattern image previously generated, then this pattern image does not need to be updated and no new pattern image is created. Still according to this standard, the following data structure can then be used for each pattern of the set 520: - object of the pattern: data structure describing a pattern (see description above), - integer: value of importance of the pattern (value initialized for example at 0), - boolean: indicates whether the pattern is marked or not (initialized to FALSE). The determination of the image patterns performed in step 206 is followed by the generation of a pattern image at step 208. The algorithm of FIG. 5 comprises a step 512 which follows step 510 and provides for generating an image from the previously marked patterns (patterns satisfying the two predetermined criteria). In the example described, it is considered that the pattern image is of the same size as the images of the video sequence. However, it may be envisaged to have images of patterns of different size according to an alternative embodiment.

The pattern image is created by grouping selected patterns and arranging them to form an image of the size of a real image of the video clip. Note that, preferably, a pattern image contains all the patterns relating to the same image when possible. If, however, the number of image patterns related to the same image is too large to be contained in a single image, it may be possible to create multiple images. In this case, step 512 is repeated for each generated image. According to an alternative embodiment, a larger image of patterns is generated than that of the real images of the video sequence. When the number of patterns selected at the end of step 510 is not sufficient to construct such an image, it is possible either to: - define an image smaller than the size of the real images 15 of the video sequence, - or repeat step 206 by increasing the value of the number N of the largest image patterns. Step 512 for generating a pattern image is followed by step 516 in which this image is marked as not to be displayed when processed by device 156 of Fig. 1b. This marking can be achieved by adding a parameter in the header slices of the image. According to one variant, this marking can be achieved by using a message separate from the image data and which indicates the number of the artificial image not to be displayed. This message is defined, for example, using the RTP protocol. According to this protocol, an RTP data packet whose payload contains the number of the tagged picture and which must not be displayed is created and sent to the client device 156. The receiving device (client) then uses this packet 30 to determine the marked images and thus differentiate these from the actual images of the video sequence and which are to be displayed. The generated pattern image is illustrated in Figure 5 by reference 518.

Note that in this image the position of an image pattern may be chosen to minimize the cost of encoding the motion vectors that will be used for motion compensation. Indeed, for example, when an image pattern is always located at the same spatial coordinates in an image, it is advisable to place it at that same position in the pattern image. In this way, the new images of the video sequence containing this pattern and which will be encoded by motion prediction from it will use, as a reference, the pattern image with the position of the pattern considered in this image.

Thus, the motion vectors assigned to the zones corresponding to this pattern will be zero or close to zero, which will greatly reduce the cost of encoding these vectors. Therefore, in order to reduce the coding cost of future images, the selected patterns are placed suitably in the image or images of patterns generated at the coordinates that average their position in the previously encoded images of the video sequence. The pattern image 518 obtained in step 516 is not encoded. However, as has been seen in the description of FIG. 2, it may or may not be encoded in the encoder carrying out step 202 before being transmitted to step 210. This pattern image uniting specific patterns according to two criteria is then stored in the temporary storage area of reference 500. Indeed, this image contains patterns considered most relevant for the motion picture prediction of the sequence.

Note also that marking information assigned to the image in step 516 is transmitted to the encoder prior to transmission. A method of encoding the pattern image 518 generated in step 208 will now be described with reference to FIG.

The algorithm of FIG. 6 details the main steps performed during the encoding step 202 of FIG. 2. The algorithm of FIG. 6 is intended to show that the pattern image is encoded from the images. of patterns and images of the video sequence present in the reference storage area. Thus, most of the patterns to be encoded are already present in the reference storage area and, therefore, the cost of encoding the pattern images is low. In practice, the encoding algorithm selects the best candidate as a reference. Several methods of encoding the pattern image will be described hereinafter according to the type of codec used. According to a first exemplary embodiment, an H.264 type codec having a long-term memory and a short-term memory is considered. Pattern images generated and transmitted to the encoder are marked as long-term references and will therefore be stored in the long-term reference memory. This marking is standardized in the H.264 standard and appears in the slice headers of the image in question. The reference temporary storage area (mmco) control commands are used to replace a long term reference image with a new pattern image. Following the encoding step 202, the encoded data sent during the step 210 are encoded images (real images of the video sequence or pattern images), as well as messages identifying the transmitted pattern images. According to another exemplary embodiment when the codec is of the MPEG4 part 2 type, the pattern image is marked as a reference. Maintaining the patterns in the reference temporary storage area is ensured by regularly sending pattern image (s) to the receiving device. The data transmitted to the receiver device in step 210 is identical to that transmitted in the context of the previously described H.264 standard. For each of the embodiments which has just been described, the encoding of the pattern image (step 202) is performed by the same encoder as that used for the encoding of the images of the video sequence. Therefore, the reference temporary storage area used already contains most of the image patterns that are to be kept in this reference area.

The algorithm of FIG. 6 comprises a first step 600 during which each macroblock MB of the pattern image, denoted Im1, is selected in order to be encoded. This step is followed by a test step 602 in which it is determined whether the considered macroblock is predicted from a previously generated Im2 pattern image. In other words, it is examined whether an image pattern already present in the reference temporary storage area can be used as a reference for the estimation of this macroblock. When a candidate is searched for motion prediction, a search area is defined in the reference image. Then, all the pixels of the macro-block is compared with an area of the image included in the search area. A defined criterion is used to select the best candidate. For example, this criterion is to choose the candidate minimizing the SAD or sum of the absolute difference (Sum of Absolute Difference) between the reference area and the pixels of the macroblock. If so, the image Im2 that contains this pattern is selected and used as a reference for the motion prediction (step 604). It should be noted that the image Im2 is for example a reference pattern image such as the image 500a of FIG. 5. During the following step 606, it is intended to encode the current MB macroblock according to a inter coding mode from the previously selected image. Returning to step 602, in the case where the current macro block is not found in a pattern already present in the reference temporary storage area, the next step 608 provides for another test. During this step, it is determined whether the current macroblock is predicted from an image Im3 of the video sequence mapped as a reference, for example, like the image 500c of FIG. 5. If so, the step 608 is followed by a step 610 in which the image Im3 is selected as a reference image and then goes to the step 606 already described where the current macro-block is coded in inter mode.

Returning to step 608, when the current macroblock is not predicted from an Im3 image of the video sequence, then the next encoding step 612 is provided. By performing steps 602 and 608 it has thus been found that no reference area for motion prediction is determined for the current macro block. Therefore, the encoding of the macro-block is performed in intra mode. When the current macroblock comes from a synthesized image pattern, the encoding of this macroblock is performed in intra mode from the synthesized image pattern. When the current macro block has been encoded in either step 606 or step 612, step 600 is performed again to process the next macro block of the pattern image. Note that in the exemplary embodiment, the pattern is defined as being of a size multiple of 16x16 which is the size of a macroblock. Thus, a macroblock is either an image pattern or a pattern part. When one considers, alternatively, that the patterns may be of any size, then a macroblock may correspond to more than one image pattern. However, in this variant the coding cost is higher than that of the exemplary embodiment. The pattern image encoded by the process of FIG. 6 is of a very small size insofar as this image is predominantly predicted with respect to data that is already contained in the reference temporary storage area. The coding cost is therefore very low and the bit rate associated with the transmission of these particular data is therefore very small. Other types of codecs can be used. According to these other types, the pattern image generated is not transmitted to the decoder. It is sufficient to transmit to the decoder information enabling it to reconstruct the temporary storage area of reference present to the encoder.

Thus, for example, specific commands are set and transmitted from the transmitting apparatus 150 of Fig. 1b to the receiving apparatus 156 to send all types of picture patterns of a pattern picture. These commands indicate the areas of the reference images corresponding to the patterns to be selected on the receiver side. For example, we could use a command such as Store (num_image, origin, width, height) where num_image is the number of the image in the reference storage area that contains the patterns to store, origin corresponds to the coordinates of the point d origin of the rectangular zone 10 corresponding to the patterns, and width and height respectively correspond to the width and the height of the considered pattern. The patterns whose pixels are present in a reference image (ie stored in reference zone 110) can be directly placed in a new memory zone of zone 110 by using the Store command. The parameters of this command are used to extract from an image already present in the temporary storage area the region corresponding to the pattern. This region is then copied into the new memory location of the reference area 110. This command is for example sent in a RTP data packet. When the pixels of the pattern are not present in the temporary memory area 110, a second command can be used. This command has as parameter all the pixels of the considered pattern. This set of pixels can be compressed by a fixed image compression algorithm for example. Thus, the data transmitted during step 210 of FIG. 2 are, in the example described, an encoded image of the video sequence as well as messages containing commands making it possible to generate the pattern image at the device level. receiver (client). Fig. 7 shows a computer or communication device 700 which may be a microcomputer or a workstation. This device is connected to different peripherals, for example a digital camera 701 for capturing video or any other device for acquiring or storing images, such as a scanner, providing the computer with videos to be transmitted. These videos can be stored in the storage means available to the computer. It will be noted that the communication devices 150 and 156 of FIG. 1 b are for example of the type of this device 700.

The computer 700 also comprises a communication interface 702 which is connected to a communication network 703, for example the Internet network, and which is capable of transmitting and receiving digital data. The computer 700 also includes data storage means such as a hard disk 704, a floppy disk drive 705 for writing data to a floppy disk 706 and reading this data. The computer may also include a compact disc player (CDROM or DVDROM) not shown, on which the videos can be stored and a computer card reader (PC-GARD) not shown. The executable codes of the program or programs making it possible to implement the different parts of the invention are, for example, stored in the hard disk 704. According to an alternative embodiment, the executable codes of these programs are stored in a ROM memory. (Read Only Memory) 707 of the computer. The computer 700 further includes a screen 708 for viewing the videos and a pointing device (not shown), such as a mouse or an optical pen, and / or a keyboard 709 so as to interact with the program. . The computer includes a central processing unit (CPU) 710, for example a microprocessor, which controls and directs the execution of the instructions of the program (s) of the invention stored in the ROM. 707 or in the hard disk 704. The computer 700 also comprises a RAM (Random Access Memory) memory 711. The latter comprises registers for storing the variables created and modified during the execution of the program or programs according to the invention, in particular the variables mentioned above in the description, in relation with FIGS. 2 to 6.

The computer 700 also includes a microphone 712 for recording sounds and storing them. The computer 700 further comprises a communication bus 713 connecting the different devices.

The programmable communication device 700 (server) comprises all the means necessary for carrying out the invention for processing data (determination of image patterns, selection of patterns according to a first criterion, identification of patterns in function of a second criterion, determination of image (s) of patterns from the grounds both selected and identified ...), encode these data and transmit them on the network. Similarly, the programmable communication device 700 (client) includes all the means necessary for the implementation of the invention to receive and process encoded video data, decode and display at least a portion. 20

Claims (20)

REVENDICATIONS1. A method of processing data of a video sequence comprising a succession of images, characterized in that the method comprises the following steps: coding (202) video images of the sequence based on a motion prediction from d reference images of the sequence, reference images being stored in a so-called reference storage area, - determining (204) image patterns each obtained from at least one constituent data set of a zone of an image, - selecting (204, 206), among the previously determined image patterns, image patterns satisfying a first predetermined criterion, - identifying (206) previously determined image patterns that satisfy a second predetermined criterion related to the reference storage area; - generation (208) of at least one image of patterns combining image patterns that satisfy both the first criterion and the second criterion; predetermined criteria; storage (208) in the reference storage area of said at least one generated pattern image. 2. Method according to claim 1, characterized in that the determination of image patterns is performed from coded images. 3. Method according to claim 1 or 2, characterized in that the first predetermined criterion reflects the importance of an image pattern for the motion picture (s) prediction of the sequence. 4. Method according to claim 3, characterized in that a pattern responds to the first predetermined criterion if it is used several times to predict the motion of images of the sequence. 5. Method according to one of claims 1 to 4, characterized in that the second predetermined criterion reflects the imminence of an image pattern to be deleted from the reference storage area or its absence from this area. 6. Method according to one of claims 1 to 5, characterized in that the number N of image patterns in said at least one pattern image is determined according to the memory capacity of the reference storage area to be used when decoding the encoded video sequence. 7. Method according to one of claims 1 to 6, characterized in that all the image patterns relating to the same image of the sequence are combined in the same pattern image. 8. Method according to one of claims 1 to 7, characterized in that it comprises a step of positioning an image pattern in said at least one pattern image so as to minimize the cost of encoding the vectors of movement. 10 9. Method according to one of claims 1 to 8, characterized in that it comprises a step of generating marking information distinguishing said at least one pattern image generated images of the sequence. 10. Method according to one of claims 1 to 9, characterized in that it comprises a step of coding (202) of said at least one pattern image 15 generated. 11. Method according to one of claims 1 to 10, characterized in that it comprises a step of transmitting (210) on a communication network of said at least one generated pattern image. 12. The method of claim 11, characterized in that it comprises a step of transmitting the coded video sequence with said at least one generated pattern image. 13. Method according to one of claims 1 to 12, characterized in that it comprises a step of long-term marking of a pattern image indicating that this image must be maintained in the reference storage area as long as the marking is present. A method of processing image data of a stream having an encoded video sequence comprising a succession of coded images according to a motion prediction based coding from reference images of the video sequence, characterized in that the method comprises the following steps: - receiving (300) image data from the stream, the image data comprising coded images and at least one image, coded or not, consisting of image patterns, each pattern image obtained from a set of data constituting an area of an image, said at least one image image image satisfying both a first predetermined criterion and a second predetermined criterion related to the reference storage area, - decoding (302) coded images in particular from said at least one pattern image for display, said at least one pattern image not being displayed. 15. Method according to claim 14, characterized in that the received stream also comprises marking information distinguishing said at least one image of image patterns from the images of the sequence. 16. The method of claim 15, characterized in that it comprises a step of selecting (304), from the received stream, images that can be displayed, the selection being performed according to the marking information. 17. Device for processing data of a video sequence comprising a succession of images, characterized in that the device comprises: means for video coding images of the sequence based on a prediction of motion from images of reference of the sequence, reference images being stored in a so-called reference storage area, means for determining image patterns, each obtained from at least one set of data constituting a zone of an image; means for selecting, from the previously determined patterns, image patterns satisfying a first predetermined criterion; means for identifying previously determined image patterns that satisfy a second predetermined criterion related to the image; reference storage area; means for generating at least one image of patterns comprising image patterns which satisfy both the first criterion and the second one; th predetermined criterion, - storage means in the reference storage area of said at least one generated pattern image. 18. An image data processing device of a stream having an encoded video sequence comprising a succession of coded images according to a motion prediction coding from reference images of the video sequence, characterized in that the device comprises: - receiving means (300) of the image data of the stream, the image data comprising coded images and at least one image, coded or not, consisting of image patterns, each image pattern; image having been obtained from a set of data constituting an area of an image, said at least one image pattern image satisfying both a first predetermined criterion and a second predetermined criterion related to the reference storage area, coded image decoding means (302) in particular from said at least one pattern image for display, said at least one pattern image not being displayed. 19. An information storage medium, possibly totally or partially removable, readable by a computer system, comprising instructions for a computer program adapted to implement the data processing method according to any one of claims 1 to 16. when the program is loaded and executed by the computer system. A microprocessor-readable computer program product comprising portions of software code adapted to implement the data processing method according to any one of claims 1 to 16, when loaded and executed by the microprocessor. .