PROCESS AND DEVICE OF COMPRESSION Of IMAGE, SYSTEM OF TELECOMMUNICATION COMPRISING SUCH a DEVICE AND PROGRAM IMPLEMENTING SUCH a PROCESS

The present invention relates to a method and an apparatus for compressing images, a telecommunication system comprising such a device and a program implementing such a method. The invention is applicable, in particular, to video compression systems capable of providing different levels of quality, in the dimension SNR (acronym standing for "A signal to noise ratio" for signal to noise ratio).

The future compression system emerging from scalable SVC (acronym standing for "Scalable Video Coding" for video coding or hierarchical), extension of the standard video compression/AVC H264, is being normalization. The aim of the new standard is to provide a compressed representation adaptable (English "scalable") or hierarchical a digital video sequence. SVC provided support a scalability, or adaptability, according to the following three axes: temporal scalability, and spatial quality.

For the scalability as, it can take two different forms in the SVC current specification. Indeed, an enhancement layer may be as CGS (acronym standing for " Coarse grain Scalability " for adaptability to large grain) or FGS (acronym standing for "fine grain Scalability " for adaptability to fine granularity).

A layer of refining CGS type contains, both, the refinement data, the motion data and texture data. A layer of CGS combines quality, on the one hand, the motion-compensated temporal prediction and, on the other hand, the predictive coding of the motion and texture data from its base layer.

A type FGS enhancement layer contains data progressive refinement texture information. One or more layers of successive FGS quality may be encoded at top of the base layer or a layer of spatial scalability or CGS. Typically, nested Quantification means and progressive encoding DCT coefficients (acronym of "or" for discrete cosine transform) provides a bit stream FGS fitted, adapted to be truncated to any position and gradually increasing the quality of the entire image considered.

Contribution technique In the JVT-P 059 presented to the union of JVT (acronym standing for "Joint Video Team" for joint video team) of Poznan, July 2005 :" Comparison of MCTF and closed-loop hierarchical B pictures ", it is shown a comparison of the coding efficiency obtained by applying the motion compensation loop open, that is between original images of the sequence to be encoded, and closed loop, i.e. by using as a reference picture images reconstructed versions in the highest rate FGS. The contribution shows that the best performance is achieved using the estimate moving closed loop.

The contribution technique JVT-P 0,057 presented to the union of JVT (joint Video Team) of Poznan, July 2005: " Implementation of closed-loop in JSVM coding " results in a similar conclusion.

However, the inventors have observed that the quality layer FGS largest for a user is not the quality layer FGS maximum but the layer effectively receives after transmission. Therefore, encoding using motion estimation taking, as a reference, a reconstruction of the reference picture of the maximum quality level, will not be optimal, to the direction of the compression efficiency, if the user receives an SVC stream at a level intermediate quality, lower than the level of maximum quality.

A optimizing the efficiency of encoding for the quality level type FGS most important for the user, for example the quality level corresponding to the level, or gap, the more demand flow by a plurality of clients at a given time.

To this end, according to a first aspect, the invention concerns a method for compressing a sequence of images, comprising, for at least a portion of a image to be compressed:

-a step of obtaining at least one parameter value representative of the operation of at least one device for decompressing compressed picture;

-a step of selecting a quality level as a function of at least one said parameter value;

-a step of estimating at least one motion vector between a portion of the image to be compressed and a portion of a reference image reconstructed by selected quality level and

-a step of coding at least said portion of image to be compressed using each estimated motion vector.

Therefore, the present invention allows for dynamic selection of the quality level of the reference images based on the request expressed by the user, to optimize the quality of the rendered image for the majority of those users.

Among other advantages of the present invention, it is observed that the use of the proposed method for video compression within the encoder, or the associated device, does not require any modification of the method and the decoding device.

According to particular characteristics, during the step of obtaining at least one parameter value, a parameter for which at least one value is representative of a flow rate used for at least one data transmission of compressed data to at least one device for decompressing compressed image.

Therefore, the present invention allows for dynamic selection of the quality level of the reference images according to the different levels of rate used by users to pressure relief devices, to optimize the quality of the rendered image for the majority of those users.

According to particular characteristics, during the step of selecting a quality level, is determined, from a plurality of ranges of values of a predetermined parameter, the the majority, at least relative, of the values of said parameter used by decompressing compressed image and selecting a corresponding quality level, in a predetermined manner, to said range of values.

According to particular characteristics, during the step of obtaining at least one parameter value, at least one parameter for which at least one value is representative of a quality level performed by a device for decompressing compressed image.

According to particular characteristics, during the step of selecting a quality level, the quality level is selected which performs rate-distortion optimized the choice of the motion vectors and reference images reconstructed used for motion estimation.

According to particular characteristics, each said portion is a image macroblock, the step of selecting the quality level being carried out individually for each macroblock of at least one picture of the sequence of pictures.

By these arrangements, the optimizing is performed by macroblock macroblock, which improves the quality of the decompressed images.

According to particular characteristics, during the encoding step, SVC encoding is carried out.

According to particular characteristics, during the step of encoding, encoding is carried out of a layer called the "base" and at least one layer of quality type of adaptability to fine granularity, or FGS.

With each of these provisions, the present invention applies to optimize compression efficiency of the SVC encoder, for the quality layers corresponding to the throughput ranges predominantly requested by the plurality of clients "multicast", i.e. receiving the same media.

For the user who receives an SVC stream at selected intermediate quality, encoding may be more optimal to the quality, to the direction of the compression efficiency, since the motion estimation then taken as a reference the version of the reference frame that is actually reconstructed at the decoder of the user.

In a second aspect, the present invention relates to a device for compressing a sequence of images, which includes means for obtaining at least one parameter value representative of the operation of at least one device for decompressing compressed image and, for at least a portion of a image to be compressed:

-a means to select a quality level as a function of at least one said parameter value;

-means for estimating at least one motion vector between a portion of the image to be compressed and a portion of a reference image reconstructed by selected quality level and

-means for encoding at least said portion of image to be compressed using each estimated motion vector.

According to particular characteristics, the means for obtaining at least one parameter value is designed so that a parameter for which it obtains at least one value is representative of a flow rate used for at least one data transmission of compressed data to at least one device for decompressing compressed image.

According to particular characteristics, the means for selecting a quality level is adapted to determine, from a plurality of ranges of values of a predetermined parameter, the the majority, at least relative, of the values of said parameter used by decompressing compressed image and selecting a level of quality corresponding, in a predetermined manner, to said range of values.

According to particular characteristics, the means for obtaining at least one parameter value is designed so that at least one parameter for which it obtains at least one value is representative of a quality level performed by a device for decompressing compressed image.

According to particular characteristics, the means for selecting a quality level is adapted to select the level of quality that performs rate-distortion optimized the choice of the motion vectors and reference images reconstructed used for motion estimation.

According to particular characteristics, each said portion is a image macroblock, the selector means being adapted to select a quality level individually for each macroblock of at least one picture of the sequence of pictures.

According to particular characteristics, the encoding means is adapted to encoding SVC.

According to particular characteristics, the encoding means is adapted to perform encoding a layer called the "base" and at least one layer of quality type of adaptability to fine granularity, or FGS.

In a third aspect, the present invention relates to a telecommunication system comprising a plurality of terminal devices connected via a telecommunication network, characterized in that it comprises at least one terminal device having a compression device as disclosed above and in accordance with at least one terminal device having a decompression device adapted to reconstruct images from the data from said compression device.

According to a fourth aspect, the present invention relates to a computer program loadable into a computer system, said program containing instructions for carrying out the method of compression such that in accordance with the above, when this program is loaded and executed by a computer system.

The advantages, goals and particular characteristics of the compression method, and of the telecommunication system of the computer program being similar to those of the device of compression such that in accordance with the above, they are not recalled herein.

Other advantages, purposes and features of the present invention shall become apparent from the description that will follow, made, explanatory purposes and in no way limiting facing the accompanying drawings, in which:

-figure 1 represents, in the form of a block scheme, a particular embodiment of a device for compressing image object of the present invention,

-figure 2 represents, schematically, an organization multilayer possible with SVC,

-figure 3 illustrates the SVC hierarchical representation of Figure 2, which have been added in layers refining type FGS,

-figure 4 represents, schematically, a conventional video decoder, typically representative of the video compression standard H264/A VC,

figure 5-represents, schematically, the insertion of decoding functions of layers of FGS refinement in the decoder shown in Figure 4,

-figure 6 represents, schematically, quality levels related display encoding and decoding an incrementing the sequence of images with quality level,

-figure 7 represents, in the form of a block diagram, an encoder of the prior art,

figure 8-represents qualities obtained after decoding, in accordance with the level of quality of the reference picture used to encode,

-figure 9 represents, in the form of a block diagram, a particular embodiment of the encoding device object of the present invention,

figure 10-represents, in the form of a logigramme , implemented steps in a particular embodiment the compression method object of the present invention and

figure 11-represents, in the form of a logigramme , implemented steps for performing one of the steps illustrated in Figure 10.

Before describe the present invention, is retrieved, below, opposite the Figures 2 to 6, the principles of a multilayer representations video sequence with scalable video coding (SVC).

In the entire description, the terms "residue" and "prediction error" denote, indifferently, the same entity. Also, the terms "encoding" and "compression" denote the same functions that apply to an image and the terms "decoding", "reconstruction" and "decompression" are equivalent to each other.

In the following, we mean by "base layer" the base layer compatible with standard H264, a layer of spatial scalability or a layer of scalability CGS.

The video compression system SVC provided hierarchies, or scalabilités , time in the dimensions, spatial and quality. The temporal scalability is achieved by providing type image "B-hierarchical" in the base layer, or by the MCTF (acronym standing for "Motion Compensated temporal filtering" for motion compensated temporal filtering), non-herein, in the layers of refinement. Scalability as The, or "SNR", exists in two forms.

SNR scalability The coarse or "CGS" is provided by encoding a layer (known in English as "layer") in which at least either a temporal decomposition image B hierarchical type, or a motion compensated temporal filtering MCTF is performed regardless of the lower layer. A layer of coarse SNR scalability is predicted from the lower layer directly.

Finally, spatial scalability is obtained by predictive coding a layer in which a motion compensated temporal filtering MCTF is performed regardless of the lower layer. Encoding an enhancement layer is similar to that of a CGS layer, except that it compresses the sequence of video images at a higher level of resolution with respect to that of the lower layer. The encoding includes, in particular, an oversampling step spatial in the two spatial dimensions (width and height) in the prediction process interlayer.

SNR scalability The fine, or adaptability to fine granularity, noted "FGS", is obtained by progressive quantization. FGS encoded The layers refinement of a given layer does that carry information texture refinement. They reuse the motion vectors carried by the base layer. In the implementation of current reference the encoder SVC, this motion estimation is made either between the original image to compress and the reference images reconstructed at their highest quality FGS (motion estimation closed-loop), or between original images (motion estimation open-loop). Therefore, the motion vector calculation, and hence the coding efficiency, are optimized for the FGS maximum quality level.

Refinement type progressive FGS provides a refinement of the values of the samples of texture representing a spatial or temporal prediction error. Refinement Note that no motion information is carried by a quality layer FGS. The motion vectors associated with each macroblock predicted time are transported by the base layer above which are added the FGS layers. That is, to reconstruct a macroblock predicted time, the motion vector used in the motion compensation by the decoder is unchanged regardless of the quality level at which the decoder operates considered.

Therefore, the encoder is in charge of generating a single motion field is used for the motion compensation in the base layer (base layer PI264, spatial, or CGS), and all of the FGS layers above the base layer.

Figure 2 shows an example of organization multilayer possible with the compression system SVC. The base layer 200 represents the sequence of images to its lowest level of spatial resolution, compressed compatibly with the standard H264/AVC. As shown in Figure 2, the base layer 200 is comprised of images of type I, P and B hierarchical.

The type image hierarchical B provide a means for generating a base layer scalable, i.e. adaptable, in the time domain. Bi They are noted, i > 1, and follow the following rule: image Bi time can be predicted from the anchor pictures, reference images of type I or P which occur as boundaries of the group processed image (English "Group of Pictures" noted "GOP"), the surrounding, and images Bj, j < i, located in the same interval anchor pictures I or P are observed that between the anchor pictures, pictures are B. are observed also that an image B1, i.e. the first picture of a sequence, cannot be predicted from the anchor pictures or P I the surrounding since there is no image < Bj with j i.

In the entire series of the description, have been limited to the case where the reference image is made of the preceding reconstructed image. However, based on the description that will follow, those skilled in the art known implement the present invention in other cases in which the one or more reference images are different from the preceding reconstructed image, particularly if a plurality of reference images is used. The scope is not limited to the latter case. The present invention also covers the case of picture list references used to the temporal prediction.

In figure 2, two layers of spatial refinement, 205 and 210, are illustrated. The first spatial enhancement layer 205 is encoded predictively relative to the base layer 200, and the second spatial enhancement layer 210 is predicted from the first spatial enhancement layer 205. A step of over-sampling spatial upsamples with a factor of two is conducted during these predictions layers, also named prediction "inter layer", so that a top layer contains images with the definitions, in each dimension, double those of the immediately lower layer.

Figure 3 illustrates the SVC hierarchical representation of Figure 2, which have been added in layers refining type FGS 300 to 325. A FGS enhancement layer is a refinement in quality of the texture information. The texture information corresponds to either an error, or residue, temporal prediction, is an error, or residue, spatial prediction, or a texture-coded "Intra", without prediction. A layer of FGS scalability type provides a refinement as texture information concerned, to the lower layer. As The refinement is progressive, i.e. the bitstream segment from the FGS encoding can be truncated at any point. The result of the truncation and remains decodable provides a representation of the set of the relevant image to a quality level increasing with the length of the bit stream decoded. Also is said that the binary string generated by the FGS encoding is "progressive quality" or "nested".

These two advantageous properties of the FGS encoding (refinement as and progressiveness of the bit stream) is obtained from the two following coding tools:

-the progressive quantization: the quantization parameter assigned to a given FGS enhancement layer is such that the quantization step applied to the DCT coefficients is divided by two to the lower layer;

--the cyclic coding the DCT coefficients of different blocks of a picture: the coding order of the DCT coefficients of an image is a function of the amplitude of the different DCT coefficients. The coefficients of greatest amplitude appear first in the bit stream. Indeed, a "passes of significance" (English "significance pass") is indicative of the significant coefficients relative to a threshold amplitude. Furthermore, a refinement enables encode refinements of amplitude values of the coefficients already encoded as significant. The macroblocks are thus more in the bitstream the order of their natural path, as in encoding the SVC other layers. Instead, the DCT coefficients of the different blocks are interleaved and their scheduling depends on their respective amplitude. The cyclic coding, is referred to as "progressive refinement", provides the property of nesting the FGS bitstream, i.e. the possibility of the truncate at any point, while remaining capable of being decoded, each additional quality layer providing an increment quality spatially covering the entire image in question.

Figures 4 and 5 illustrate how supported layer processing SVC type refinement FGS within a video decoding algorithm. Figure 4 illustrates a conventional video decoder 400, typically representative of the video compression standard H264/AVC. Such a decoder includes, in a known manner, the application on each successive macroblock functions entropy decoding, functional block 405, inverse quantization, functional block 410, inverse transformation, functional block 415. Information waste of these first three operations is then added to a reference macroblock for its spatial or temporal prediction. The image exiting this prediction also passes a filter for reducing the effects of block (English " deblocking filter") 420 reducing the block effects. The image thus reconstructed is adapted to be displayed, on the one hand, and to be stored in a list 450 of reference images, on the other hand. It is, in effect, made to serve as reference image for the temporal prediction, functional block 425, of the next pictures to be decoded for the compressed bit stream, the image resulting from the temporal prediction 425 is added to the image from the inverse transform 415 by a summer 435.

Figure 5 illustrates the insertion of the decoding functions refinement layers FGS in a decoder 500 having all functions of the decoder 400 shown in Figure 4. As shown in Figure 5, the decoding layers-type progressive refinement FGS, functional blocks 505, 510 and 515, is between the function of inverse quantization 410 and the inverse transform function 415, and is successively applied to all macroblocks of the current picture being decoded.

Decoding FGS provides, over the entire picture, a refinement of the values of the samples after inverse quantization. Therefore, as shown in Figure 5, the decoding FGS provides a progressive refinement of the prediction error spatial or temporal. The refined prediction error is then passed through the same functions as in the decoder 400 of Figure 4.

Refinement type progressive FGS provides a refinement of the values of the samples of texture representing a spatial or temporal prediction error. Refinement are observed that no motion information is carried by a quality layer FGS. The motion vectors associated with each macroblock predicted time are transported by the base layer above which are added the FGS layers. That is, to reconstruct a macroblock predicted time, the motion vector used in the motion compensation by the decoder is unchanged regardless of the quality level at which the decoder operates considered.

Therefore, the encoder is in charge of generating a single motion field is used for the motion compensation in the base layer (base layer H264/AVC, spatial, or CGS), and all of the FGS layers above the base layer.

Figure 6 represents the interdependencies between the different layers FGS of the different images of a GOP (acronym of "group of pictures" for group of pictures) given SVC in a video stream. Figure 6 illustrates, on the one hand, a base layer 605, which represents a SVC layer spatial scalability, CGS or the base layer compatible H264/AVC. Images of the base layer are noted l₀ ^base , B_n ^base and P^base , in which the index n represents the picture index, the exponent base indicates the layer to which the image belongs, and I, P or B represent the type of the image. Furthermore, refinement layers FGS 610, 615 and 620, 625 as well as the original images are also illustrated in Figure 6.

The images of these FGS layers are noted 1 ∩ , Β ∩ and Ρ ∩ , in which the index scores n represents the picture index, the exponent/indicates the FGS layer to which the image belongs, and I, P or B represent the type of the image

During the process of temporal prediction macroblocks image P or B, the encoder performs motion estimation. If, the example encoding an image p₈ ^base as shown in Figure 6, the motion estimation provides, for each macroblock of the picture P₈ ^base , a motion vector the binder to a macroblock of the image l₀ ³ , i.e. the reference image reconstructed at the highest quality level. The motion vector is then used in the step of motion compensation to generate a macroblock prediction error, or residual macroblock also called residue. The macroblock residual is then encoded by quantization, and entropy coding transformation. Furthermore, the image n is encoded by refining the quantization applied to the macroblocks of any remaining image P₈ ¹ , before a cyclic coding is performed.

Several strategies can be used by the encoder for motion estimation used, without modification to the decoding algorithm. The following strategies have been scanned within the standardisation committee SVC:

-estimating open loop movement includes estimating, for each macroblock of an original image to be encoded, a motion vector between the macroblock and a macroblock a reference image in its original version. Estimating open loop movement processability is between original images of the sequence to be compressed;

-the closed loop motion estimation includes estimating motion vectors between an original image and a reconstructed version of the reference picture used. In contributions at standardizing techniques committee SVC, it is proposed to use the reference image reconstructed in the highest quality FGS for performing the estimation motion closed loop.

Studies show a better performance are obtained by performing the motion estimation closed loop, between the original image and encoding the image (s) (s) reference decoded (s) in the highest rate FGS. Indeed, work in closed loop take into account distortions introduced when the quantization reference images.

Are observed, further, that one of these contributions leads to the conclusion that the best compression performance are obtained by performing the motion compensation also closed loop to the encoder. The motion compensation closed-loop consists in calculating the macroblocks time prediction error by calculating the difference between a macroblock original to be coded and the reconstructed macroblock reference to the same level of quality FGS. This configuration of the FGS encoder leads to the better performance for all quality levels FGS.

The present invention relates primarily to the motion estimation process closed loop. The inventors have observed that the motion estimation made taking into account the reconstructed version of the original image in the highest quality FGS leads to performance optimization for the compression layer FGS the highest quality. Indeed, the motion estimation then taken into account distortions introduced in the reference image during the compression of the latter. By employing the reconstructed versions images of reference to the highest data rate FGS involves that the encoder takes into account distortions introduced when all the layers FGS is decoded.

To realize the motion estimation from reference images reconstructed to intermediate levels to optimize the encoding for these intermediate quality levels. The implementation of the present invention can select a level of quality, one of the levels of base quality and FGS, as reconstruction level images of reference to perform the motion estimation, in particular closed loop.

In embodiments of the present invention, the selection of the quality level used for motion estimation is performed according to a relative importance value assigned to each of the quality levels which can be output by the encoder. For example, in the preferred embodiment of the invention, the importance value is defined based on the proportion of clients, at each instant, each quality layer FGS during a multi-point video transmission.

Preferably, are selected dynamic of a quality level FGS for reconstruction of reference frames used then to estimate the motion vectors, according to the relative importance of the quality FGS in the transmission.

Note that changing dynamically the quality level for the reconstruction of the reference images needs not to modify the algorithm video decoding. This is unchanged, regardless of the motion estimation strategy used on the side of the encoder.

Are observed, in Figure 1, a device 100 object of the present invention, or encoder, and different devices adapted to implement the present invention. In the embodiment shown in Figure 1, the device 100 is a microcomputer of known type connected, via a graphics card 104, to a means of acquiring or storing images 101, for example a digital camera or a scanner, adapted to provide information of moving picture compressing.

The device 100 comprises a communication interface 118 connected to a network 134 adapted to transmit, input, digital data to be compressed or, output, compressed data by the device. The device 100 also includes a storage means 112, such as a hard disk, a floppy disk reader 114 and 116. The diskette 116 and the storage means 112 may contain data to be compressed, compressed data and a computer program adapted to implement the method of the present invention.

Alternatively, the program enabling the device to implement the present invention is stored in read-only memory ROM (acronym standing for "read only memory" for non-rewritable memory) 106. Alternatively, the program is received via the communication network 134 before being stored.

The device 100 is connected to a microphone 124 through a card input/output 122 for associating audio data to the picture data to be encoded. 100 This same device has a screen 108 for viewing the data to decompress (a customer's case) or interface with the user to customize some embodiments of the device 100, through a keyboard 110 and/or such as a mouse.

A CPU (acronym standing for "central processing unit") 103 executes the instructions of the computer program and program necessary for its operation, for example an operating system. Upon power-up of the device 100, the programs stored in a non-volatile memory, for example the read-only memory 106, the hard disk or floppy disk 112 116, are transferred in a random access memory (acronym standing for "random access memory" for random access memory) 105 which will contain then the executable code of the program implementing the method of the present invention and a register for storing the variables necessary for its implementation.

Of course, the disk 116 can be replaced by any removable media, such as compact disc, memory card or key. More general, information storage means, readable by a computer or by a microprocessor, integrated or not integrated with the device, possibly removable, stores a program implementing the encoding method object of the present invention. A communication bus 102 allows communication between the different elements included in the device 100 or connected to it. The representation, in Figure 1, the bus 102 is not limited and such as the central processing unit 103 is able to communicate instructions to any element of the device 100, directly or through another element of the device 100.

For the execution of the program implementing the method of the present invention, the central unit 103 performs the functions illustrated in Figure 9 and the steps illustrated in Figures 10 and 11 and constitutes the following means:

-means for obtaining at least one parameter value representative of the operation of at least one device for decompressing compressed image

-and, for at least a portion of a image to be compressed, herein each of the macroblocks image compressing:

-a means to select a quality level as a function of at least one said parameter value;

-means for estimating at least one motion vector between a portion of the image to be compressed and a portion of a reference image reconstructed by selected quality level and

-means for encoding at least said portion of image to be compressed using each estimated motion vector.

In particular embodiments, the encoding means is adapted encoding SVC-encoded FGS type quality layers. In embodiments, the selection means determines the relative importance of different levels of flow, by determining rate are most of the users or by determining a median value or an average of the levels for operation by the users, optionally using a weighted average, each level of flow rate and/or each user having a relative weight, for example in connection with a difference in distortion between implementations of different levels of quality to reconstruct the reference images. Alternatively, use is made of a cost function representing the loss of quality corresponding to a choice or other quality level reconstructed image to determine motion vectors is searched and the minimum of the cost function, it being understood that the users may not have, all, the same influence on the cost function.

The functional diagram of Figure 7 is the during, encoder side, the decryption algorithm illustrated in figure 5. Are observed, in Figure 7, a video encoder 700 FGS generating quality levels according to the state of the art. The video encoder 700 includes a video input providing image sequences, to compress, a transformation function 705, a quantization function 710 functions and three progressive refinement FGS 715 to 725, respectively for the levels to FGS3 FGS1. The progressive refinement texture data maximum quality, from the function of progressive refinement FGS 3,725, is used by a function of inverse quantization 730, followed by an inverse transform function 735, for reconstructing an image prediction error or residual at maximum quality.

The progressive refinement texture data maximum quality, from the function of progressive refinement FGS3 725, is provided, on the one hand, to an entropy coder 745, which provides, as output, the compressed images encoded.

The reference image, from the switch 750 is summed to the reconstructed residual image and transmitted to a filter for reducing the effects of block (English " deblocking filter") 740. The reconstructed image resulting filter 740 is the image recreated current version in its final, ready-to-display. The reconstructed image is further stored in a list reference picture 770.

The reference image stored in the memory space 770 is implemented by a motion estimation function 765 which determines, for each macroblock of the current picture, a motion vector and the provides, on the one hand, to the entropy encoder 745 and, on the other hand, to a compensation function 760 that uses motion, further, the reference image from the memory 770.

The step of motion compensation macroblock 760 provides a reference for the temporal prediction of each macroblock of the current picture. Furthermore, the step of intra-frame prediction 755 determines, for each block of the current macroblock being processed, a reference block for its spatial prediction. The role of the switch 750 is then choose the mode of coding, from the temporal prediction, and the spatial prediction encoding INTRA, which provides the best performance for the current compression macroblock. The mode selection to the direction rate-distortion optimized thus provides the reference macroblock used to predict each macroblock of the current picture. As a result a prediction picture of the current picture. As shown in Figure 7, the difference between the original image and image current prediction is calculated, and is the image prediction error encoding. It is operated by the steps of transforming, quantification and entropy encoding mentioned previously.

Therefore, the video encoder 700 generates a base layer and a plurality of layers of progressive refinement FGS above the base layer. The block diagram of Figure 7 illustrates a video encoder typically standard H264/AVC, wherein the generating functions QoS levels type FGS 715 to 725 have been added. FGS These refinements come progressively increase the quantization of the base layer, by dividing by two the quantization step of a quality level FGS relative to the previous quality level. The quantization indices of transformed coefficients in the base layer, and the elements refinement quantization FGS layers are provided to the entropy encoder 745 of generating the compressed bit stream scalable SNR in the dimension.

In parallel, a reconstruction is carried out by the functions 730 to 740, to form a reference image which is used for estimation and for the motion compensation performed by the functions 760 and 765.

Figure 8 shows a benefit of implementing the present invention, compression performance. Are observed, in Figure 8, the different curves rate-distortion 805, 810, 815 820 and it is contemplated that obtain when the motion estimation is done by successively using the different levels of base quality and FGS can be delivered by the encoder. On each of these curves, lower distortion, represented in ordered, and the greater the quality of the image. Figure 8 illustrates in that take, as references for the motion estimation, the reconstructed images to a given quality level leads to optimization coding for the range of flow rates corresponding to the quality.

For example, select the FGS maximum quality level, herein FGS3, to reconstruct the reference images for the motion estimation closed-loop corresponds to a rate distortion curve 820 below the other curves to 815 805, i.e. a reconstructed image of better quality, for the range of rate corresponding precisely to the quality, right of the Figure.

Furthermore, Figure 8 shows a hypothetical histogram 825 for different values of rate being actually received by a set of clients in a shaft multicast transmission, these rate values being values representative of the operation of the client devices. It appears, on this example, that the range of the most important, i.e. the more "requested" by the client population, corresponds to a range of data rate compatible with the second level of quality FGS, called FGS2.

By with certain embodiments of the present invention, the optimization of the SVC for the encoding quality level.

In other embodiments of the present invention, the optimization of the SVC for the encoding quality level corresponding to a minimum of a cost function representing the corresponding loss of quality, by all users, the selection of a level of quality reconstructed image to determine motion vectors.

Note that the principle of the invention is also applicable in the practical case of a point-to-point video transmission, i.e. from a video server to a single client. In this case, the range of flow rates or large corresponds to the relevant rate being actually received by the unique client. The bandwidth corresponds to a quality layer FGS given type. According to the present invention, coding performance is optimized for the quality FGS, and thus the motion estimation using as reference images of the reconstructed images precisely to the quality used by the client.

Therefore, in accordance with the present invention, adjustment of the level of quality of reconstruction of reference pictures for the motion estimation based on at least one value of at least one parameter representative of the operation of at least one device for decompressing compressed image, for example the throughput values or QoS levels used for decompression.

Are observed, in Figure 9, a block diagram of a particular embodiment of an FGS encoder 900 using the present invention. Since the video encoder 700 shown in Figure 7, the video encoder 900 generates a base layer, compatible H264/AVC, and layers of progressive refinement type FGS, as a function of a quality level selected. Is retrieved thus, in Figure 9, the same functional blocks that in the encoder shown in Figure 7. However, to the functional blocks, is added a mechanism 905 adaptive choice of the level of quality FGS to which the reference images are reconstructed for the motion estimation in closed loop, in accordance with the level of quality of maximum importance.

The mechanism 905, represented as a switch transmitting the quantized transformed coefficients in one of the four possible quality levels (base, FGS1, or FGS2 FGS3) function of inverse quantization 730, takes into account information from the transmission network and indicating the amount of clients each of the layers of quality of the base layer and the layers of FGS refinement in the embodiment described herein. Generally, the information from the network contain values of parameters representative of the operation of the client devices, capable of receiving and expanding the compressed image.

For example, a return mechanism customer information to the encoder gathers the throughput values received by the customers connected to said network. The server associated with the video encoder 900 is capable of determining the flow ranges corresponding to each of the levels of quality delivered by the encoder and transmitted to the clients. For example, using the teaching of the document "Text of ISO/IEC 14496 Advanced Video Coding 3rd Edition" of g. Sullivan, Wiegand t. and A. Luthra, available from ISO/IEC/JTCs 1/SC 29/WG 11, Redmond, WA, USA, is mapped between the lengths of the NAL (acronym standing for "Network With Layer", for network abstraction layer) units, or transfer units of the bit stream, corresponding to each quality layer and flow rates indicated by the returnmessages from the network. The mechanism is described lower, opposite Figure 11.

This mapping allows the encoder determining proportion of each receiving client quality levels available at the output of the encoder and transmitted by the video server. This ratio of clients is used to define the relative importance of each quality layer generated by the video encoder. The relative importance is used to perform the selection of the quality level or base FGS for the reconstruction of images of reference within the temporal prediction loop implemented by the inverse quantization functions and inverse transformation of video compression. Therefore, the encoder 900 uses, as a reference picture, in its motion estimator 765, the reconstructed images and displayed by a majority, at least relative, of the contemplated application multicast. This improves subsequent quality video seen by the majority of clients.

Figure 10 represents, in the form of a logigramme , the following steps are performed in a particular embodiment the method of the present invention, to complete the encoding of a sequence of images, with a base layer and one or more layers of progressive refinement above the base layer.

In a step 1005, receives an original image is compressing, as well as information from relative importance of each quality level, is calculated and provided by the method shown in Figure 11.

In step 1005, for each of the original image macroblock current, Estimation of movement after desired, in a manner known per se, in a reference image, a macroblock resembling most the sense of a rate-distortion criterion. The found macroblock macroblock and serves as a reference for the temporal prediction macroblock of the original current. The difference between the two macroblocks is the prediction error signal, which is compressed via the steps of transforming 1012, quantization, step 1015, and entropy coding, step 1055.

To form the layers of FGS refinement, the quantization step 1015 est followed by a plurality of successive quantization with a quantization step divided by two quality levels between two successive FGS, in a step 1020. The result of these successive quantization is implemented during the entropy encoding step 1055 to generate a bit stream representing the video sequence in compressed form.

Furthermore, each macroblock prediction error and compressed is then reconstructed. To this is first a step of inverse quantization 1025. The inverse quantization is carried out at maximum quality of relative importance determined by the method, in the form of logigramme , in Figure 11. In step 1025, inverse quantization is progressively applied to the image up to the level of the maximum quality of relative importance. Furthermore, the transformed coefficients obtained after inverse quantization, step 1025, are inverse transformed, step 1030. Each macroblock prediction error is added to its thus reconstructed macroblock reference, step 1035, to provide a reconstructed macroblock. These steps being applied to each macroblock of the picture, the current picture is thus completely reconstructed by quality level of maximum importance. The reconstructed image is then reducing the effects of filtering block 1037 (deblocking filter), and then stored in a reference picture list, in a step 1040.

In a step 1045, it is determined whether the processed image corresponds to the last of the sequence of images to be encoded. If yes, aborting said method, step 1060. Otherwise, in a step 1050, is passed to the next image of the sequence of images to be encoded and returning to step 1005. The image stored reconstructed by selected quality level serves as a reference picture for the motion estimation applied to future images to be encoded.

The step of previously detailed reconstruction is therefore performed such that the motion estimation for the next images of the sequence is carried out with reference to images reconstructed at most important quality, for example the level predominantly perceived by customers.

Figure 11 represents, in the form of a logigramme , implemented steps for selecting the quality level of maximum relative importance, of the base layer and one of the layers of FGS type delivered by the video encoder considered.

In a step 1105, information is obtained from the network, on the rates received by the client population of the shaft considered multicast transmission. In the particular embodiment described herein, this information takes the form of a number of clients receiving a given flow rate. The assembly rates is quantized and reduced to a limited number of intervals achievable rate. The information returned by the network is imaged by a table of numbers of clients Nbdients for each flow rate index k, noted rate Rk, of the set of possible rates. Note that mechanisms exist to retrieve that information synthesizing the reception conditions for each customer, and are not detailed herein.

The following steps illustrated in Figure 11, to compute the values of relative importance for each level of quality q from the group {base, FGS1, FGS2, FGS3 \. In the embodiment the method of the present invention shown in Figures 10 and 11, the importance of each quality level is defined as the proportion of clients that receive the quality level considered. The importance is first initialized to 0 for each level of quality in a step 1110. In a step 1115, for each flow rate Rk, with quality levels (or base FGS) generated by the video encoder and delivered by the video server, the amount of information is calculated for each quality level delivered by the server per unit time, on a sliding time window. This amount of information is calculated by summing the lengths of the units NAL (or transfer unit of the bit stream SVC) transmitted by the video server on the duration of the time window. NAL These lengths of units are known by the video server, since the units are precisely NAL generated and transmitted via this video server. This amount of calculated information providing a flow emitted for each level of quality. Is then determined, for a rate Rk given, the layer of the highest quality, from the base layer, when rate value, in a step 1117. This is given by:

r fgs₃

Q = Arg min j ℮ Μ Σΐ ^^)≥ R_k

qe {base, FGS_t , FGS₂ , FGS₃ } [= base q

wherein length (q) nAL represents the total length of the units output for the quality level q. That is, the value of the flow rate Rk received by certain clients provides a number of quality levels from the base level.

In a step 1120, for the level of quality affected by the maximum rate value Rk, is updated the value of relative importance of the quality. This update takes the following form:

I_Q <- I_Q + NbClients\R_k ]

Indeed, the magnitude of the highest quality level concerned Q is increased as there are clients that receive the rate Rk.

In a step 1125, it is determined whether Rk is the last rate interval to be considered. If yes, is passed to step 1135. Otherwise, in a step 1130, is next rate interval and returning to step 1115.

In step 1135, each importance value are normalized by dividing by the sum of the significance values calculated. This allows for the relative importance value between 0 and 1 for each flow rate Rk. Finally, the quality level of greater relative importance is selected in a step 1140.

The level the most important of which is then taken into account from step 1025, as shown in Figure 10, for reconstruction of the reference picture.

The sequence of the description introduced another particular embodiment the method of the present invention. The inputs of this embodiment consist of the different intervals of flow rates Rk multipoint received by the plurality of clients. Is then determined the range of the higher data rate, i.e. which corresponds to a throughput received by a majority of clients. The rate of maximum importance simple is thus determined by the following expression:

R= max R\_k \.

The rate value R of maximum importance one of the plurality of rates received by the plurality of clients is then taken into account in the motion estimation algorithm contained in the prediction process the video encoder time considered. In the particular embodiment described herein, the motion estimation process utilizes an optimization algorithm rate distortion, known to those skilled in the art and included in the reference software SVC, to estimate the motion vectors binder the block of the current picture to be coded to their reference blocks.

The thereby alters the rate distortion optimization algorithm implemented in the software reference SVC, called JSVM (acronym standing for "gasket Scalable Video Model" model scalable video unified), which provide a common reference software to the members of the JVT committee for evaluating performance of pressing tools provided by members of the group. Indeed, for each partition, sub-macroblock of a partition P of a macroblock image B to be encoded, the motion estimation consists of searching a reference block in a reference picture which minimizes the Lagrangian according to:

^m o/i(^r o/i)^- A-^r g min \l)_SAI) (P_j , r_0/l , /"_0/l )+ / i _SAI) [/ " (r)+ Κ (μο ∩ ) J | (1) m_is eù

where the distortion D_S ad, for a partition of macroblock or sub-macroblock P, is given by the following expression:

^ S AD 1'/ ^m o / 1) Σ I j]^_ re / ^. O/Lt^m 0/l,x ' J^m 0/l, }-] I fê) (UhP

In the equation 2, l_orig represents the set of samples of the original image being encoded and l_re f,o/i represents the samples of the reference picture used for the search for the best predictor of the current macroblock. The symbol 0/1 models the fact that the search is carried out in sequence on the lists indexed "0" and "1" of reference images, the list of index "0" containing the reference images in the past (L₀ ), used for prediction before, and the list of index "1" containing images future, used for prediction (L-i) rear. In the equation (1), S is the search space for the motion vectors. The terms R (R₀ / i) and R (M₀ / i) specify the cost (number of bits) coding-related indexes r₀ / i and of the motion vector components m₀ / i.

Once generated candidate motion vectors for each partition sub-macroblock Pi, i is the index of partition sub-macroblock macroblock in the partition P, in each of the reference images of the lists L₀ and L_h one selects the reference images r₀ in R₀ and n in Ru and the associated motion vectors m₀ and m? that minimize lagrangian according to:

^r o/i ⁼ Arg min \ ' y^_i (D_SAn (P_i , r_is , m_is (r_n/l ,?)))+ sl_SAn x R (m_is (r_is , i) ') + At_{SAD CALCULATOR} xi? (r_0/1 ) l

^r o/i^e ^o/i [^ ieP J

eL/

(3)

For infeed of the notion of relative importance of each level of quality in the FGS selection mechanisms, is modifies the definition of the distortion measure D_S ad, as indicated by the equation (4) below.

^D sad calculator ( P >^r 0/i > level, rn_is ) = level (level) x £I l_orig [i, j]-l_reffi ,_Uevel [* '+ m_0/lx , j + m_0/ly ] I (Uhr

(4)

wherein The level (level) &[ 0, l] represents a measure of relative importance calculated by performing the steps illustrated in Figure 11.

The level (level) is measured for each level of quality level in L = {base, fgsi , fgs2 , fgs3 }. Therefore, l_re f, O/ ijevei represents the set of samples of a reference picture candidate reconstructed by quality level level. Finally, the last step of selecting the reference image, according to the equation (3), is also modified. Indeed, it includes, in addition, the selection of the quality level is decoded used to which the reference picture used for the partitioning of current macroblock P. This selection step takes to present the form of the equation (5):

(r_is , level) = Arg min lΣ (ΰ ^ (Ρ ; , r_is , level, m_is (r_is , i))) + At_{SAD CALCULATOR} x R (m_is (r_is , i)) + At_{SAD CALCULATOR} χ R (r_is ) ^r 0/l^e ^0/l I i < = P

1 < = L

(5)

Therefore, rate-distortion optimization the choice of the motion vectors and reference images reconstructed used to estimate

closed loop motion is performed. The embodiment of the present invention produces superior results to the previous, performance point of view of compression, since the effective content reference images reconstructed at each of the levels of quality FGS is taken into account.

Furthermore, in this embodiment, the selection of the quality level FGS the reference block to the motion estimation is performed adaptively for each macroblock of the current picture during compression.

Therefore, the motion estimation process is performed using as reference image (s) one or more reconstructed images

at quality FGS selected according to practical conditions of transmission, for example the bandwidth, multipoint in a given environment. The perceived video quality is optimized for a flow rate, or level of quality, required by a majority, at least relative, of clients.

Therefore, the context convenient flow transmitting the different values typically scalable bandwidth available in the network multicast considered-is taken into consideration for determining the relative importance of a quality layer FGS from a plurality of layers

fGS quality delivered by the encoder SVC.

The practice of the present invention makes it possible to dynamically optimizing compression efficiency of the SVC encoder for the quality layers corresponding actual needs of the various clients multicast.

Therefore, the present invention provides the functionality of progressive encoding the texture information and applies, in particular, in the case of the SVC system being standardization, but also to any encoder having the ability to encode representative samples of a signal gradually and nested, or hierarchical, such as by using quantization techniques and nested coding by bit planes.

Note that the use of the method or device of the present invention, at the encoder, does not require any modification of the system or decoding method.