FR2929790A1 - Data e.g. scalable video coding image, transmitting method for client, involves transmitting layer of priorities at required switching during reception of request for switching between initial and target spatial resolutions from client - Google Patents

Abstract

The method involves determining switching between two levels of different spatial resolution, and generating a scalability layer of priorities at the level of spatial resolution in the absence of switching. The scalability layer of priorities is generated in the switching, and the scalability layer of priorities is transmitted at the required switching during reception of a request for switching between the level of initial spatial resolution and a level of target spatial resolution from a part of client. Independent claims are also included for the following: (1) a device for transmitting data between a server and a client (2) a computer program comprising a set of instructions for implementing a method for transmitting data between a server and a client (3) a information medium comprising a set of instructions for implementing a method for transmitting data between a server and a client.

Description

The present invention relates to a method and a device for transmitting data. It applies, in particular, to the transmission of adaptable video streams (in English scalable) SVC (acronym for scalable video coding for adaptable video coding) containing several levels of spatial resolution, from a server to a client, with the possibility of modify the adaptability or scalability layers transmitted during the video session, in particular the level of spatial resolution received by the client. In a client-server SVC encoding system, it may be interesting for the user to navigate between scalability layers, especially between spatial layers. The aim here is to render a decoded video sequence with the best quality possible on the client side when switching between spatial scalability layers. For this purpose, it is necessary for the video decoding process to be able to switch from one spatial level to another at certain locations in the video sequence.

Several technical solutions already exist for this. Drift reduction techniques are known during switching operations, in particular the introduction of so-called IDR images (acronym for Instantaneous Decoder Refresh), the introduction of so-called key SVC images, the H. images. 264 SP and SVC ESP proposed to the JVT (acronym for joint video team for video team), the technique described in the French patent application No. 06 50974 filed March 21, 2006 and the technical solution described in US 6,961,383. These different techniques are analyzed below. A first SVC video coding technique for switching between spatial resolution levels is to frequently insert intra-coded images (i.e. without reference to other images) in the different spatial layers. This solution has the disadvantage of an additional cost in terms of compression ratio associated with intra coding. A second video coding SVC technique for switching between spatial resolution levels is to use so-called key images. These images, encoded in a base layer of a given spatial enhancement layer, are systematically completely reconstructed during the decoding process and are used as reference images of the future images in the enhancement layer. Therefore, this coding technique leads to greater complexity, compared with a single loop decode mode as well as a loss of coding efficiency in the enhancement layer. French Patent Application No. 06 50974 describes a coding method which comprises, when switching from a so-called current image, an initial quality level to a target quality level different from the initial quality level. and reconstructing the current image to the desired quality level to provide a coded current image, obtaining each reference image of the current image at the initial quality level, determining a differential texture refinement prediction image function for the initial quality level and the reconstructed current target quality level image and a differential texture refinement code step. This technique therefore has the disadvantage of increasing the amount of information to be transmitted. The problem underlying the present invention is to render the best possible video quality to a given client during a switching requested by the customer, between two different levels of spatial resolution. For this purpose, the present invention aims, in a first aspect, a method of data transmission between a server and a client, characterized in that it comprises, performed by said server: - a step of determining at least one possible switching between two different spatial resolution levels, - a step of generating at least one priority layer adapted to a spatial resolution level, in the absence of switching, - a step of generating at least one layer of a priority adapted to a said possible switching, and - when receiving a switching request between an initial spatial resolution level and a target spatial resolution level, on the part of the client, a transmission step of the layer of priority adapted to the requested switching. This takes advantage of the phenomenon of time drift when switching between two spatial levels SVC, at least when this drift provides a gain in quality compared to a drift drift. A time drift in the decoder is thus allowed if it improves the visual quality. The advantages of the present invention include: - to provide the customer with the best quality according to the bit rates available for this customer, at the time of a change of spatial resolution level and - there is no cost of implementation of the invention, in terms of bit rate, since no additional information is transmitted during the switching. According to particular characteristics, during the step of generating at least one priority layer adapted to a said possible switching, a group of decoded images obtained during a switching between the initial spatial level and the level is generated. target space by resampling the last image from a previous group of images to fit the target spatial level.

According to particular characteristics, during the step of generating at least one priority layer adapted to a said possible switching, the images of the current image group are decoded at the target spatial level with, as a reference image in the temporal prediction, the decoded and resampled image.

According to particular features, during the step of generating at least one priority layer adapted to a said possible switching, the images of the current image group thus decoded are calculated with the distortions related to the decoding of elementary entities of the bit stream of the current image group. According to particular characteristics, during the step of generating at least one priority layer adapted to a said possible switching, each priority layer is constituted according to the ratios of the distortions related to the decoding of each elementary entity of the bit stream. current image group by the rate necessary to transmit said elementary entity. According to particular characteristics, during the step of generating at least one priority layer adapted to a said possible switching, a bit rate-distortion optimization is performed to assign priority layer indices, or priority levels, to each elementary entity of the current group of images. Thus, a modified set of elementary entities ("NAL units") is transmitted to the client where the level of spatial resolution is modified in comparison with the approaches known in the prior art. According to particular features, the step of generating at least one priority layer adapted to a said possible switching comprises a storage step, during which an identifier of at least one priority layer defined is stored according to said priority layer indices. According to particular features, during the storage step, SEI (Supplemental Enhancement Information) messages are written which encode and store useful data for optimal sub-stream extraction and transmission, when the client or the network causes a spatial switch. The present invention aims, according to a second aspect, a device for transmitting data between a server and a client, characterized in that it comprises, in a server: a means for determining at least one possible switching between two levels of different spatial resolution, - means for generating at least one priority layer adapted to a spatial resolution level, in the absence of switching, - means for generating at least one priority layer adapted to a said switching possible, and - a suitable transmission means, when receiving a switching request between an initial spatial resolution level and a target spatial resolution level, on the part of the client, to transmit the adapted priority layer at the requested switching. According to a third aspect, the present invention provides a computer program loadable in a computer system, said program containing instructions for implementing the transmission method object of the present invention as briefly described above. According to a fourth aspect, the present invention aims at a support of information readable by a computer or a microprocessor, removable or not, retaining instructions of a computer program, characterized in that it allows the implementation of the transmission method object of the present invention as succinctly set forth above. The advantages, aims and special features of this device, this program and this information medium being similar to those of the method as succinctly set forth above, they are not recalled here. Other advantages, aims and particular features of the present invention will emerge from the description which follows, made for an explanatory and non-limiting purpose, with reference to the accompanying drawings, in which: FIG. 1 represents, schematically, a particular embodiment of the device object of the present invention; FIG. 2 represents a drift when switching from one spatial resolution level to another; FIGS. 3A and 3B show, schematically, priority layers implemented in an SVC encoding and their representation in a curve; FIG. 4 schematically represents the processing of priority layers with a view to the spatial resolution level switching and FIGS. 5 to 8 represent, in the form of a logic diagram, steps implemented in a particular embodiment of FIG. process object of the present invention. FIG. 1 shows that, in a particular embodiment, the device that is the subject of the present invention takes the form of a microcomputer 100 equipped with software implementing the method that is the subject of the present invention and different peripheral devices. The device constitutes, here a server adapted to transmit coded images to customers (not shown). The microcomputer 100 is connected to different peripherals, for example a means of acquiring or storing images 107, for example a digital camera or a scanner, connected to a graphics card (not shown) and providing information about image to compress and transmit. The microcomputer 100 includes a communication interface 118 connected to a network 134 capable of transmitting digital data to compress and transmit data compressed by the microcomputer. The microcomputer 100 also includes a storage means 112 such as, for example, a hard disk. The microcomputer 100 also includes a floppy disk drive 114. A floppy disk 116, such as the storage means 108, may contain compressed or compressed data. The diskette 116 may also contain instructions of a software implementing the method that is the subject of the present invention, instructions which, when read by the microcomputer 100, are stored in the storage means 108. According to one variant, the program enabling the device implementing the present invention is stored in ROM 104 (called ROM, acronym for read only memory, in FIG. 1). In the second variant, the program is received via the communication network 134 and is stored in the storage means 108. The microcomputer 100 is connected to a microphone 124 via the input / output card 122. The microcomputer 100 has a screen 108 for viewing the data to be compressed or to interface with the user, using a keyboard 110 or any other means (mouse for example). Of course, the floppy disk 116 may be replaced by any information medium such as CD-ROM (acronym for compact disc read only memory for read-only memory compact disk) or a memory card. More generally, an information storage means, readable by a computer or by a microprocessor, whether or not integrated into the device, possibly removable, stores a program implementing the method that is the subject of the present invention.

A CPU 120 (called CPU, acronym for central processing unit, in Figure 1) executes the instructions of the software implementing the method object of the present invention. When powering on, the programs enabling the implementation of the method that is the subject of the present invention stored in a non-volatile memory, for example the ROM 104, are transferred into RAM RAM 106 which then contains the instructions of this software as well as registers to memorize the variables necessary for the implementation of the invention. A communication bus 102 allows communication between the different elements of the microcomputer 100 or connected to it. The representation of the bus 102 is not limiting. In particular, the central unit 120 is capable of communicating instructions to any element of the device directly or via another element of the device. Figure 2 illustrates the image quality that can be obtained by decoding an SVC sequence when switching from a higher spatial resolution level to a lower spatial resolution level. The horizontal lines 205 and 210 represent the quality levels, in terms of signal-to-noise ratio, for example PSNR (acronym for Peak Signal to Noise Ratio for peak signal-to-noise ratio), obtained when the levels of the signal are regularly decoded. higher and lower spatial resolution, respectively, without performing level switching.

Accordingly, if a switch is performed without any drift by the decoder, the image quality obtained by the user corresponds to the line 215, during switching. On the contrary, if a time drift is present during the downward switching from the higher spatial resolution level to the lower spatial resolution level, the resulting image quality 220 can be between the two lines 205 and 210. The present invention takes advantage of this quality of image, which is superior to that obtained without temporal drift.

FIGS. 3A and 3B illustrate the principle of the priority layers implemented in the adaptable (scalable) video coding SVC. In Figure 3A is shown an image group, or GOP of a video sequence. In FIG. 3A, only two levels of spatial resolution 305 and 310 are represented, it being understood that more than two levels can be implemented in the SVC coding. In addition, each spatial level contains two quality levels, 306 and 307, for the spatial level 305, and 311 and 312, for the spatial level 310. In each spatial level, each image of the GOP is identified by its type of coding, P or B, and for the first, the index of the GOP and, for seconds, its index in the sequence. Thus, successively, the images P0, BO, B1, B2, B3, B4, B5, B6 and P1. Finally, since the temporal adaptability is provided by the SVC encoding by encoding hierarchical B-images, each GOP image has a time-level identifier, indicated above the images. Here this identifier successively takes the values t0, t3, t2, t3, t1, t3, t2, t3 and t0. Images with the highest temporal levels can be excluded to extract a sub-stream (substream) with a reduced temporal resolution, or frame rate. In the following, each image of a spatial level is identified by its quality level and its identifier in the quality level. For example, B3, Q0 corresponds to the image B having an index 3 and a quality level Q0, in the spatial level considered. In terms of NAL elements (Network Abstraction Layer for Network Abstraction Layer) units.de bitstream, such an image is encoded in one or more elements, depending on the number of macroblock slices (slices ) used to encode this image (one NAL unit per macroblock slice). A NAL unit represents an elementary H.264 or SVC data entity, and consists of a header and a body). The header provides characteristics of the data contained in the body. The body contains either data of an encoded slice or encoding parameters of the sequence or images of the video sequence. Here, for the sake of clarity, we do not consider the possibility of encoding auxiliary images, which would carry redundant data in a dedicated NAL unit that would be used to increase the robustness of the transmission, facing transmission errors. Figure 3B illustrates how the priority layers are calculated for the GOP images illustrated in Figure 3A for a given spatial level. In general, the calculation of the priority layers aims to provide each coded NAL unit contained in the GOP at the given spatial level a so-called priority ID value value. This priority identifier therefore represents the contribution of each NAL unit to the quality of the decoded version of the GOP considered, according to the cost of the NAL unit, in terms of bit rate. Thus, the NAL units that make the greatest contribution to the quality of the reconstructed GOP get the highest priority level and highest transmission priority. In other words, when the server must select a substream for a target rate, the substream is formed by first selecting the NAL units with the highest priority levels. Thus, the sub-stream having the best video quality is formed by the substream extraction agent ("substream extraction agent") as a function of the required bit rate, in the following it explains how the priority layers are calculated in SVC A priority layer consists of all NAL units that have the same priority identifier value in an SVC bit stream.

The determination of the priority layers is carried out in three main phases: firstly, the flow-distortion information of each NAL unit is calculated. The bit rate associated with each NAL unit constituting the GOP is given by the length, in bytes, of the encoded NAL unit. The distortion is obtained by calculating the distortion reduction obtained by decoding the GOP with or without the NAL unit under consideration. More details on the different ways of calculating the distortion associated with a given NAL unit are given in the article by I. Amonou, N. Gammas, S. Kervadec, and S. Pateux. Optimized rate-distortion extraction scalable extension of h.264 / avc published in IEEE Transactions on Circuits and Systems for Video Technology, 17 (9): 1186-1193, September 2007. - then the rate-distortion curve 320 is determined in the coding of the current GOP, as illustrated in FIG. 3B. This rate-distortion curve 320 is obtained by sorting the NAL units according to their rate-distortion slope. This slope is obtained by dividing the distortion reduction associated with the NAL unit by the cost in terms of flow rate of this NAL unit. Thus, the NAL units are sorted in descending order of their slope, which provides the rate-distortion curve 320 associated with the coding of the current GOP in the spatial level considered. It is noted that constraints imposed by the temporal dependence between the images of the GOP have been taken into account so that reference data appear before the corresponding predicted data, in the list of NAL units. Section III.B of the document mentioned in the previous paragraph provides precise algorithmic elements to ensure compliance with these constraints. and finally, the priority layers (called glittering in the figure) are formed as a function of the ordered flow-distortion points and each desired layer rate. This phase brings together adjacent NAL units at the end of the previous phase into quality levels. Different criteria can be used to decide how to join adjacent NAL units in the same priority layer. For example, a target bit rate required for each priority layer can serve as a criterion. The priority layer identifiers are encoded in the SVC bit stream either in the header in each NAL unit, in a six-bit field called priority_id, or in the SEI message called "Priority layer information" (for layer information priority). The syntax of this message is described in the article by T. Wiegand, G. Sullivan, J. Reichel, H. Schwarz, and M. Wien Joint Draft ITU-T Rec. H.264 / ISO / IEC 14496-10 / Amd.3 Scalable Video Coding. Joint Video Team (JVT) of ISO / IEC MPEG & ITU-T VCEG, Geneva, Switzerland, July 2007. Document JVTX201, section G.13.6. Once the priority layers are calculated and written in the SVC bit stream, they can be used in the sub-stream extraction process. The priority layers indicate optimal rate-distortion decoding points in the SVC bit stream. Thus, when an SVC server is to transmit a substream at a given rate, it can use the priority layer information to retrieve the priority layers that provide a substream with a rate equal to the required rate. The technical advantage of using priority layers to perform this extraction is that the extracted sub-stream provides the best video quality for the required bit rate. Thus, the priority layers ensure optimal sub-stream extraction and transmission in client / server SVC systems. FIG. 4 illustrates the calculation of the priority layers dedicated to spatial switching, in order to exploit the quality improvement that can be obtained by the time drift during a spatial level switching contained in an SVC video sequence. In the embodiment of the present invention illustrated in FIG. 4, for each GOP of the sequence, one or two additional sets of priority layer identifiers are computed which are associated with possible spatial level switches for the considered GOP. An example of an SVC sequence with three spatial levels 480, 481 and 482 is shown in FIG. 4. A right horizontal arrow represents the regular decoding of a GOP at the intermediate spatial level. A down arrow represents the decoding path followed when switching from the higher spatial level 482 to the intermediate spatial level 481. An upward arrow represents the decoding path followed when switching from the lower spatial level 480 to the level The PLSEI references of the SEI messages of the NAL units of the priority layers are given below the images of the intermediate spatial level 481. They comprise, in the first line, the priority layer identifiers normally calculated as described above with respect to Figure 3. For each intermediate spatial level GOP image, two other priority layer SEI messages are shown in Figure 4. The first, written just below the PLSEI reference, denoted PLSEI_do, represents the quality level information. which is calculated taking into account the downward switching between the GOP previous at the higher 482 spatial level at the intermediate GOP 481. The second, written just below the reference PLSEI_do, denoted PLSEI_up represents the quality level information which is calculated taking into account the upward switching between the GOP previous lower spatial level 480 to the intermediate GOP at the intermediate spatial level 481. The advantage of calculating additional sets of priority layers and writing them in the bitstream is that this additional priority layer information can be used by the server SVC when the client performs a spatial level switch. The SVC server can then use the pre-recorded priority layer information that is adapted to the switching made by the client. For this purpose, the SVC server extracts and transmits the sub-stream which presents the optimal bit rate-distortion performance for the considered switching since the priority layers used are precisely calculated to determine a decoding configuration which is optimal in bit rate-distortion for this particular switching. The phases performed to calculate the priority layers adapted to the commutation considered are described below. It is noted that the calculation of priority layers is performed offline (off-line) by the SVC server and that its result is preserved in the bit stream to be transmitted. Thus, the cost, in terms of processing, related to the calculation of the priority layers has no impact on the transmission process in real time. For each decoding path associated with each possible switch (ascending or descending), for each GOP, for each spatial level (spatial level identifier in the video stream dependency_id) and for each image representation in said spatial level (dependency unit which represents the set of NAL Units corresponding to an image in a given spatial level), we replace the image I or P of the previous GOP by its decoded version in the starting spatial level resampled at the spatial level of arrival (or target) of the commutation considered. The conventional priority layer assignment process is performed on the obtained GOP. An additional Priority Layer information SEI message, or priority layer information in French, corresponding to the spatial level switching considered, is generated to the current spatial level (481 in FIG. 4). Each server-side priority layer information is recorded by distinguishing the information from the regular SEI messages and the new priority layer identifiers by generating a proprietary priority layer SEI message which signals the priority layer identifier adapted for switching. . According to H.264 / AVC and SVC specifications, such proprietary SEI messages are ignored by decoders that do not know them. Then, when the real-time transmission of the SVC bit stream thus increased takes place in the server, in the event of switching between two spatial levels, the server achieves optimal extraction in the sense-distortion direction through the priority layer SEI messages. recorded and corresponding to the switch in question. Since these layers have been determined by taking into account the decoding process implemented by the client during the commutation in question, they necessarily give debit-distortion performances greater than or equal to those provided by the conventional priority layers (identified by the PLSEI messages in Figure 4) that do not take into account spatial switching.

It is noted that the new priority layers thus reported in these proprietary SEI messages are not transmitted to the client. They are only used at the server level to determine the best set of NAL units to transmit in case of spatial level switching.

FIG. 5 represents a global algorithm for calculating priority layers in a particular embodiment of the method that is the subject of the present invention. The objective of this algorithm is to assign three different indices, for the intermediate layers (and two, for extreme layers), layer of priority optimized in terms of rate-distortion at each NAL unit of the bit stream SVC: - the standard priority layer index corresponding to the decoding of an SVC sub-stream without spatial level change during decoding, - the debit-distortion optimized priority layer index in the context of a spatial switch from a lower spatial level to the current spatial level and - the bias-optimized priority layer index in the context of spatial switching from a higher spatial level to the current spatial level. As a variant, two different indices for the intermediate layers, and one index for the extreme layers, of a rate-distortion optimized priority layer for each NAL unit of the SVC bitstream, including the standard priority layer, are attributed. This algorithm corresponds to the SVC standard priority layer calculation algorithm described in the article by J. Reichel, H. Schwarz, and M.

Wien Joint Scalable Video Model JSVM-11 ISO / IEC JTC1 / SC29 / WG11 and ITU-T SG16 Q.6, Geneva, Switzerland, July 2007, Joint Video Team document JVT-X202, to which are added the determinations of priority assigned to different NAL units in the context of upward or downward switching between spatial levels.

The inputs of the algorithm illustrated in FIG. 5 are the following: the bit stream B for which it is sought to calculate the priority layers, the number NbDid of spatial resolution levels contained in the bit stream B, the number of quality levels contained in each level of spatial resolution of the considered SVC bit stream: {NbQL (d), V of [0, ..., NbDid - 1]}. According to this algorithm, a loop is performed comprising steps 510 to 540 on the successive spatial levels d contained in bit stream B. After a step 505 for initializing the current spatial level d to 0, for each spatial level d, it is performed a step 510 which consists in calculating the cost in terms of rate 6R (d, q, t, i) and the decrease in distortion 6D (d, q, t, i) of each NAL unit of the bit stream B of spatial level d . This calculation is identical to that carried out in the standard algorithm for estimating the SVC priority layers described in the document cited in the preceding paragraph and is known to those skilled in the art. Indeed, it corresponds to the calculation of bit rate and distortion for a decoding of an SVC sub-stream without spatial level change during decoding. In the above notations, the quantities q and t respectively represent the quality and time levels to which the NAL unit contributes. The index i is the index of the current image in the spatial level, of quality and time considered. In the case where each image is coded in a single slice of macroblock, a single NAL unit containing encoded image data is present for each index combination (d, q, t, i). The rest of the algorithm consists in calculating distortions associated with each NAL unit of spatial level d in the context of commutations, or commutations, between spatial levels. During a step 515, it is determined whether the spatial level d is strictly greater than 0. In the case where d is strictly greater than 0, an upward spatial switching, from a spatial level below d to the spatial level of is possible since there is at least one spatial level below level d. Then, during a step 520, the distortions related to the decoding of the NAL units are calculated in the context of a spatial level change at the GOP (group of images) in which the NAL unit in question is located. This step 520 is detailed with regard to FIG. 6. If the result of step 515 is negative, that is to say if the spatial level d is equal to 0, or following step 520, during a step 525, it is determined whether the spatial level d is strictly less than its maximum value NbDid-1. If yes, a downward switch from spatial level higher than d to level d is possible. During a step 530, the distortions associated with the decoding of each NAL unit are calculated, in the case where a downward spatial switch is performed at the level of the GOP in which the current NAL unit is located. This step 530 is detailed with respect to FIG. 7. If the result of step 525 is negative or following step 530, that is, once the bit rate and the three types of distortion associated with each NAL units of the current spatial level are obtained during a step 535, the optimal optimal priority layers are calculated, that is to say considering a decoding without spatial change, then the priority layers in the context of upward and downward spatial commutations. This step is detailed with reference to FIG. 8. Then, during a step 540, it is determined whether the level d is strictly less than its maximum value NbDid-1. If yes, the level d is incremented and then we return to step 510. Otherwise, the algorithm illustrated in FIG. 5 ends because the priority layers have been calculated for each spatial level d contained in the SVC B bit stream. FIG. 6 illustrates the distortion calculation algorithm associated with the decoding of the NAL units of a given spatial level d of the considered bitstream B, in the context of an upward spatial level switching, from the spatial level d-1, directly less than d, towards level d. The inputs of the algorithm are the following: the bit stream in process B, the spatial level of current for which the algorithm of FIG. 5 seeks to optimize the priority layers in terms of bit rate-distortion and the number of quality levels contained in bit stream B at spatial resolution level d: NbQL (d). The algorithm operates a loop (steps 610 to 650) on the quality levels contained in the current spatial level d of bitstream B. For each quality level, a loop (steps 615 to 640) on the bitstream B images spatial level d and quality level q is covered. I (d, q, t, i) is the commonly processed image. In the following, we also speak of quality increment because the bitstream portion corresponding to these parameters, spatial, quality and time, constitute refinement data in SVC quality. The indices t and i respectively denote the temporal level of the image considered and the index of the image within the levels, spatial, quality, and temporal to which it belongs. During a step 605, the current quality level q is initialized to the value O. Then, during a step 610, an initialization of the current image I (d, q, t, i) to the first image of the substream of parameters d and q. During a step 615, it is determined whether the temporal level of the current image is equal to 0. If yes, step 655 is followed, during which the current image is decoded at the lower spatial level. Then, during a step 660, the decoded image is oversampled during the step 655, so as to generate a decoded spatial resolution image equal to that of the spatial level d. The result of this decoding followed by the oversampling is then stored in a temporary image denoted Iref (d, q, t = 0, i) and will be used later for calculating the distortions associated with the images of the GOP according to the GOP containing the current image I (d, q, t, i). During a step 665, the temporal level 0 reference image preceding the current GOP is replaced by the image resulting from the decoding followed by spatial oversampling performed on the preceding GOP: Iref (d, q, t = 0, i - 1). With reference to FIG. 4, this step amounts to replacing the image of the type P noted S1, Q0 in FIG. 4 by the image of the spatial level SO of the same time instant in its decoded and oversampled version towards the spatial level S1. . Thus, the type 1 or P image that serves as a reference for the temporal prediction of the current GOP images is modified, so as to simulate the behavior obtained at the decoder (client side) during a spatial level change being made. decoding of the SVC sequence considered. It is considered here that the spatial level changes are always made at the time level images O. This change of reference image implemented during a spatial level change therefore has an impact on the decoding of all the images. the next zero time level GOP corresponding to the instant of the spatial level change. The objective of the algorithm of FIG. 6 is precisely to determine the impact of this spatial switching on the distortion of the decoded images and to provide the distortion values thus obtained to the optimization process of the priority layers. Once the reference image replacement of the current GOP has been performed, step 620 is performed. If the result of step 615 is negative or following step 665, in step 620, the current image is decoded without the NAL unit containing the quality increment I (d, q, t, i). During a step 625, the current image is decoded with the NAL unit containing the quality increment I (d, q, t, i). The decrease in distortion, corresponding to an increase in quality, related to the decoding of the current image with the quality increment I (d, q, t, i) is then calculated, during a step 630, and is denoted 6Dup (d, q, t, i). During a step 635, it is determined whether the image I (d, q, t, i) is the last image of the sub stream of parameters d and q. If not, during a step 640, the next image of the sub-stream under consideration is passed to the next image and step 615 is returned. If the result of step 635 is positive, in step 645 , we determine if q is the last level of quality in the current spatial level d. If no, the value of q is incremented during a step 650 and step 610 is returned. If the result of step 645 is positive, the algorithm is completed. The algorithm of FIG. 6 thus ends when all the quality levels contained in the spatial level d of the bit stream B have been processed. Figure 7 illustrates the distortion calculation in the context of a downward spatial switch. The algorithm of FIG. 7 is similar to that of FIG. 6 and aims to calculate the distortion values associated with the NAL units of bit stream B, when these are decoded as part of a downward switching between two levels. space, occurring during a video session. The same treatments as in the algorithm of FIG. 6 are implemented, but are adapted to the case of a change from a spatial level higher than the current level d towards the current spatial level d. Thus, the steps 705 to 725 and 735 to 750 are respectively identical to the steps 605 to 625 and 635 to 650. During the step 730, the decrease in distortion, corresponding to an increase in quality, related to the decoding of the current image with the quality increment I (d, q, t, i) is computed and is denoted 6Ddown (d, q, t, i). If the result of step 715 is positive, in step 755, the current image is decoded at the higher spatial level. Then, during a step 760, the decoded image is sub-sampled during step 755, so as to generate a decoded spatial resolution image equal to that of the spatial level d. The result of this decoding followed by the subsampling is then stored in a temporary image denoted Iref (d, q, t = 0, i) and will be used later for calculating the distortions associated with the images of the GOP according to the GOP containing the current image I (d, q, t, i). During a step 765, the temporal level 0 reference image preceding the current GOP is replaced by the image resulting from the decoding followed by spatial subsampling performed on the preceding GOP: Iref (d, q, t = 0, i - 1). FIG. 8 illustrates the determination of the indexes of optimal priority layers in a particular embodiment of the method that is the subject of the present invention. This FIG. 8 details the step of calculating the standard priority layers adapted to the spatial level changes of the algorithm illustrated in FIG. 5. The inputs of the algorithm of FIG. 8 are the following: the current bit stream B processing, - the current spatial level d in which the priority layers are sought to be optimized, - the number of NbQL quality levels (d) contained in the current spatial level and - the number of temporal levels in each quality level. current spatial level: {NbTL (d, q), V q E [0,. ,,, NbQL (d) - 1]}.

The algorithm illustrated in FIG. 8 traverses a loop (steps 810 to 840) on the different levels of quality contained in the current spatial level d. For each quality level q, a temporal scalability level index t is initialized at the highest temporal level contained in the quality level q, during a step 810 preceded by a step 805 during which the level of quality current q is initialized to the value O. Following step 810, during a step 815, for each index image i of the maximum temporal scalability level in the current quality and spatial levels, one initializes a list of quality increments denoted LOd, q, t, i. The increments in this list correspond to the NAL unit containing the coded data for the current image in the current scalability layer. Three (or two for the extreme layers) increments constituting this list, are denoted I (d, q, t, i), Iup (d, q, t, i) and IdQWn (d, q, t, i), and correspond respectively to the decoding of the NAL unit considered without spatial level change, with upward switching of spatial level and with downward switching of spatial level. These three increments are ordered in the list LOd, q, t, i, according to their respective flow-distortion slope, 6D (d, q, t, i) / 6R (d, q, t, i), 6Dup ( d, q, t, i) / 6Rup (d, q, t, i) and 6DdoWn (d, q, t, i) / 6Rdown (d, q, t, i). The rest of the algorithm is known to those skilled in the art and consists in traversing all the temporal levels from the time level directly below the highest level to the time level O. Thus, during a step 820, we go to the just lower temporal level, by decreasing the current time level by 1. Then, during a step 825, it is determined whether the temporal level is greater than or equal to 0. If yes, during a step 830, for each image of the time level considered, a merger and a sorting of the lists of quality increments of images of the higher temporal level and dependent on the current image, denoted LOd, q, ft, fi, so as to form the ordered list of quality increments associated with the current image: LOd, q, t, i. This loop on the temporal levels from the highest to the lowest provides finally, for the level of quality currently treated, a list of merged and ordered quality increments denoted LOd, q, 0,0. Then we return to step 820. If the result of step 825 is negative, during a step 835, it is determined whether the quality level q is equal to the number NbQL (d) -1. If not, during a step 840, the value of the quality level q is incremented and we return to step 810. If the result of step 835 is positive, it means that a list of increments of Merged and ordered quality was obtained for each quality level q contained in the spatial level d. During a step 845, the lists associated with each quality level are merged together. This step is also known to those skilled in the art. Its realization consists of gathering and ordering the different quality increments of the different lists by decreasing rate decreasing distortion, while respecting the dependencies between reference images and predicted images, imposing to position the reference images in front of the images that depend on them. The unique list of resulting quality increments is denoted LOd, 0,0,0.

Then, during a step 850, the priority layer indices are assigned to the different spatial level NAL units d of the bitstream B, in a standard way, ie considering only sub-stream decodings. at constant spatial level during decoding. This allocation therefore ignores the presence in the ordered list of quality increments of type Iup (d, q, t, i) and Idown (d, q, t, i), which only concern the cases of upward and downward spatial commutations. . This step 850 of assigning priority layer indices is performed identically to the standard method mentioned above. Then, during a step 855, priority layer indices are assigned to the spatial level NAL units d of the bit stream B, taking into account the case where these would be decoded during a spatial level change. . This step 855 uses the scheduling of the quality increments corresponding to the upward and downward spatial switches, with respect to the standard quality increments, within the list LOd, 0,0,0 in which they coexist. The increments corresponding to spatial switches are assigned a priority layer index equal to that of the neighboring standard increments in the ordered common list LOd, 0,0,0. In this way, in an SVC video server, during a spatial level change during a video session, the priority layer indices corresponding to the current spatial switching, and attributed to the NAL units contained in the GOP at at which spatial change takes place, will be taken into account in the choice of NAL units to be transmitted in order to provide the best possible video quality to the customer during the spatial switching. For example, with reference to FIG. 4, if a downward spatial switch from the highest spatial layer to the intermediate spatial layer occurs at the first GOP illustrated in FIG. 4, the intermediate priority layer indices stored in the messages SEI rated "PLEI do" will be taken into account by the server in the SVC sub-stream extraction process to be transmitted to the client. Finally, the last step of the algorithm of FIG. 8, step 860, consists in storing the priority layer indices attributed to the NAL units and corresponding to the different spatial switches, for example in messages of the SEI type specifically defined for this need. . Such SEI type messages are therefore non-standard and can only be used by a server that would know these messages. However, this does not pose an interoperability problem since these messages only serve to improve the SVC sub-stream extraction process in the case of spatial switching, and are not transmitted to the SVC clients.

Claims (11)

REVENDICATIONS1. A method of data transmission between a server and a client (100), characterized in that it comprises, realized by said server: - a step (515, 525) of determining at least one possible switching between two levels of resolution different spatial steps, - a step (535, 850) for generating at least one priority layer adapted to a spatial resolution level, in the absence of switching, - a generation step (520, 530, 535, 855) at least one priority layer adapted to a said possible switching, and - when receiving a switching request between an initial spatial resolution level and a target spatial resolution level, on the part of the client, a step of transmitting the priority layer adapted to the requested switching. 2. Method according to claim 1, characterized in that, during the step (520, 530, 535, 855) of generating at least one priority layer adapted to a said possible switching, a group of decoded images obtained when switching between the initial spatial level and the target spatial level by resampling (660, 760) the last image of a previous group of images to adapt it to the target spatial level. 3. Method according to claim 2, characterized in that, during the step (520, 530, 535, 855) of generating at least one priority layer adapted to a said possible switching, is decoded (655, 755) the images of the group of images running at the target spatial level with, as reference image in the temporal prediction, the decoded and resampled image. 4. Method according to claim 3, characterized in that, during the step (520, 530, 535, 855) of generating at least one layer of priority adapted to a said possible switching, is calculated (630, 730), with the images of the current image group thus decoded, the distortions related to the decoding of elementary entities of the bitstream of the current image group. 5. Method according to claim 4, characterized in that, during the step (520, 530, 535, 855) of generating at least one priority layer adapted to a said possible switching, constitutes (855) each priority layer according to the ratios of the distortions related to the decoding of each elementary entity of the bit stream of the current image group by the bit rate necessary to transmit said elementary entity. 6. Method according to any one of claims 2 to 5, characterized in that, during the step (520, 530, 535, 855) of generating at least one layer of priority adapted to a said possible switching a rate-distortion optimization is performed (855) for assigning priority layer indices, or priority levels, to each elementary entity of the current image group. 7. Method according to claim 6, characterized in that the step (520, 530, 535, 855) of generating at least one priority layer adapted to a said possible switching comprises a step (860) of storage, at the wherein at least one priority layer is stored in accordance with said priority layer indices. 8. Method according to claim 7, characterized in that during the step (860) of storage, SEI messages (acronym for Supplemental Enhancement Information) are written which code and store data useful for an extraction and transmission of data. optimal substreams, when the client or the network causes a spatial switch. 9. Device (100) for data transmission between a server and a client, characterized in that it comprises, in a server: - means (104, 106, 120) for determining at least one possible switching between two different spatial resolution levels, - means (104, 106, 120) for generating at least one priority layer adapted to a spatial resolution level, in the absence of switching, - means (104, 106, 120 ) of generating at least one priority layer adapted to a said possible switching, and a means (104, 106, 120) of adapted transmission, when receiving a switching request between a spatial resolution level initial and a target spatial resolution level, on the part of the client, to transmit the priority layer adapted to the requested switching. Computer program loadable into a computer system (100), said program containing instructions for carrying out the method of any one of claims 1 to 8. 11. Support (116, 112) of information readable by a computer (100) or a microprocessor (120), removable or not, retaining instructions of a computer program, characterized in that it allows the implementation of the process according to any one of claims 1 to 8.