ITTO20100311A1 - "PROCEDURE FOR CODIFYING / DECODING VIDEO / IMAGE SIGNALS WITH MULTIPLE DESCRIPTION AND ITS CODIFICATION / DECODING APPROACH" - Google Patents

Description

DESCRIPTION of the industrial invention entitled:

â € œCoding / decoding process of video / image signals with multiple description and related coding / decoding apparatusâ €

TEXT OF THE DESCRIPTION

The present invention relates to coding / decoding techniques of source of video / image signals with multiple description, in particular techniques of image signals of the scalable type, in particular for transmission on channels with packet transmission of telecommunications networks, which includes to operate on a sequence of digital images organized in frames a layered coding on said sequence in order to obtain a plurality of layers comprising a Base Layer with a first level of quality and one or more Enhancement Layers to which respective higher levels of quality are associated.

Source coding means a mapping from a sequence of symbols generated by a source to a sequence of symbols of an alphabet (usually binary), such that the source symbols can be exactly recovered from the bits (lossless encoding) or recovered with some distortion (lossy coding). This is the concept behind data compression.

In other words, it can be said that it is the way in which each event associated with a discrete source is encoded by means of a sequence of symbols.

In general, in the following of this description we will refer to the digital images, or frames, of a sequence of images with the term â € frameâ € ™, commonly adopted in the main digital coding techniques, such as JPEG and MPEG, to define the € ™ complete image and derived images to create coding schemes such as inter-frame (P-frame) and intra-frame (I-frame).

The Moving Pictures Experts Group (MPEG) has provided scalable video coding schemes to meet different levels of picture quality depending on broadcast environments and user-level terminal performance and, together with ITU-T recommendations, has proposed a advanced video coding standard for broadcasting or broadcasting and internet multimedia services. In the scalable coding techniques (with SNR scalability, temporal or spatial), adopted in the MPEG-2 format, the compressed bit stream is divided into a Base Layer (BL), or base layer and one or more Enhancement Layers ( EL), or enrichment layer. The Base Layer guarantees the basic quality of the encoded video, while an improvement in quality can be obtained by receiving the Enhancement Layer in addition to the Base Layer bitstream. The main weakness of such scalable layered encoding is that the Enhancement Layer must be fully transmitted / received / decoded or does not provide any enhancements.

To address this limitation of previous scalable video encoding standards, Fine Granularity Scalable (FGS) video encoding has been proposed within the MPEG-4 committee, as outlined for example in publication W. Li, â € œOverview of fine granularity scalability in MPEG-4 video standard, â € IEEE Trans. on Circuits and Syst. for Video Tech., vol. 11, no. 3, pp. 301â € “317, Mar 2001. The FGS encoding, in addition to the Base Layer, includes a single Enhancement Layer encoded progressively. Even if the Base layer is encoded at a respective certain RBL rate, such that the minimum received bandwidth is greater than RBL (RBL <Rmin), the Enhancement Layer can be compressed to any desired bit-rate range [Rmin ; Rmax]. Thanks to the progressive nature of the Enhancement Layer, the decoder is able to reconstruct the Enhancement Layer for each received bit rate value. While in scalable video encoding the Enhancement Layer is entirely received or lost, with FGS encoding any part of the Enhancement Layer preceding a possible reception error improves the quality of the decoded video stream.

However, the trade-off between coding efficiency and robustness is still an open question. In particular, most of the latest generation video codecs incorporate motion-compensated prediction (MCP), leading to higher compression of the video source. With MCP prediction, a single-bit error can cause the so-called drift problem, which is the deterioration of quality, due to a propagation error due to the prediction mismatch between encoder and decoder. In some MPEG-4 FGS encoders, MCP prediction is considered for the Base Layer but not for the Enhancement Layer, avoiding the problem of drift when the Enhancement Layer is lost. It is known to improve coding efficiency by integrating the Enhancement Layer into the MCP cycle.

Solutions known in the literature (for example, progressive FGS, lossy level video encoding) lead to an encoded stream which has higher encoding efficiency, but which can suffer from error propagation when the MCP prediction of the Enhancement Layer is lost.

The object of the present invention is that of solving the drawbacks of the known art and, in particular, of realizing an encoding process capable of generating an embedded bitstream robust with respect to packet loss.

According to the present invention, this object is achieved thanks to a process having the characteristics referred to specifically in the following claims, which form an integral part of the present description.

The invention also relates to the corresponding coding / decoding apparatus, as well as the corresponding computer product that can be directly loaded into the memory of a digital computer such as a processor and comprising portions of software code to implement the process according to the invention when the product is It is performed on a computer.

According to an important aspect of the invention, the process involves operating a layered coding on a sequence of frames to obtain a plurality of layers, including a Base Layer with a first level of quality and one or more enrichment layers with respective higher levels of quality, the process further comprising of encoding the multiple descriptions of each of said layers, dividing each of said layers into a matrix of blocks, each independent of each other, in said matrix of blocks of the layer select blocks for transmission through the steps of calculating for each current block of said layer, at block level, an estimator of difference with respect to a corresponding block of a previous frame and comparing said estimator with a settable threshold to evaluate whether to transmit the current block .

Thanks to the characteristics indicated above, the process according to the invention allows to obtain an encoder apparatus with a high degree of scalability capable of dynamically satisfying the bandwidth requirements and which, compared to known scalable video encoders, has better performance when the packet loss on the channel is greater than zero.

Further characteristics and advantages of the invention will emerge from the following description with reference to the attached drawings, provided purely by way of non-limiting example, in which:

- Figure 1 schematically shows a sequence of frames used by the process according to the invention,

- figure 2 is a block diagram representing a coding apparatus according to the invention, - figure 3 is a block diagram of a module of the coding apparatus of figure 2,

- figure 4 represents a block diagram of a layered coding module of the coding apparatus of figure 2;

figure 5 represents a block diagram of a decoding module according to the invention;

- figure 6 shows an illustrative diagram of the performances achieved by the process according to the invention,

- figure 7 is a schematic representation of an image signal processed by the process according to the invention,

In short, a coding procedure and a coding apparatus are described in which the progressive source stream is combined with a multiple description type coding at the application layer, rather than at the physical layer.

In the coding with multiple descriptions, known in general for example from the publication â € œV. K. Goyal, â € œMultiple description coding: Compression meets the networkâ € IEEE Signal Processing Mag., Vol. 18, no. 5, pp.

74â € “93, Sept. 2001, the bit stream is fragmented into a plurality of descriptions, leading to more reliable image transmission over error-prone wireless channels. Since each description can be decoded independently, the information bits can be recovered despite packet loss. The quality of the decoded video stream depends on the number of successfully received packets. Due to the sensitivity to errors of progressive encoding, the first studies dealt with protecting packets from error by exploiting the MD concept at the channel encoder level, for example through forward error correction (FEC).

It is envisaged to use a source encoder suitable for encoding a video sequence or a sequence of frames in packet streams of the layered or layered type, fragmented into independent descriptions. A further level of scalability is achieved by transmitting only a subset of the encoded blocks. This means that several levels of Quality of Service, or Quality of Service (QoS) can be achieved through a single coded frame, by adopting intelligent selection of the transmitted blocks. Although the source coding technique according to the invention in the embodiment described here is based on the Joint Photographic Experts Group (JPEG) standard, for which reference can be made to the CCITT Recommendation ITU-T T.81 (09/92) , in particular Chapter IV, p. 12-22, it can be extended to other standards, which operate compression coding, such as, for example, Moving Pictures Experts Group (MPEG), Set Partitioning In Hierarchical Trees (SPIHT), and other similar standards. Other forms of encoding, for example of the wavelet type, are also compatible with the proposed solution. The preferred embodiment described here refers to the JPEG standard for its flexibility and because it requires limited computing power, suitable for embedded systems and sensor networks.

The proposed source coding procedure is now described, with reference to a JPEG codec for a sequence of frames (or video), taking into account the objectives of scalability and robustness to packet loss. Figure 1 schematically shows a possible sequence S of digital image signals, structured in frames, used for example by the method according to the invention, which comprises a number of frames Nf, as shown in Figure 1. With FI therefore indicated an intraframe or I-frame, encoded without any reference , while FP indicates an inter-frame or P-frame encoded using the previous I-frame FI as a reference. The sequence of image signals S shown comprises an I-frame every Nf-1 P-frame. The number Nf of frames can assume different values, in the examples shown here it is 12. The proposed procedure provides that both for the intra-frames FI and for the inter-frames FP of the sequence S, in the first place the image conveyed by them is subjected to a procedure of layered coding, also called layered coding, which produces a number L of layers, or layers, of the type intra- or inter- respectively, depending on whether they are derived from intra-frame FI or inter-frame FP, for each complete frame. These layers are indicated with I <0>, I <1,> â € ¦ I <l>, I <L>, for I-

frame or P <0>, P <1,> â € ¦ P <l>, P <L> for P-frames, with the integer index indicating the l-th layer among the L layers. Subsequently, it is expected that each layer in said plurality of layers is coded in a scalable way.

In this regard, Figure 2 indicates a coding system according to the invention, which receives the sequence of digital images S at the input which is coded by a layered encoder module according to a layered coding procedure. Each layer, I <l> or P <l>, is then sent to a scalable multi-description encoding module 12, which includes 13 block subdivision modules and a selector module 14, set with aging parameters and adjustable threshold for select the blocks to be transmitted on the channel.

The layered encoding procedure that is applied in layered encoder 11 is now described.

The proposed encoder allows to encode each YUV frame, where YUV 4: 2: 0 indicates the color space of each frame, in the L layer I <l> or P <l>, each of which provides a decoded image at quality better than the previous layer.

When we consider an original I-frame FI, the respective layer I <0> of order 0, that is the Base Layer, is obtained by encoding this original frame FI after a low pass filtering and subsampling, or downsampling, which leads to obtaining a low quality with high coding efficiency. In particular, denoting with a filtered and subsampled I-frame, the JPEG encoded Base Layer I <0> is:

I <0> = JPG (F <0>

I) (1) Note that the original k-th I-frame FI should be denoted by FI, k, but for simplicity k, time frame sequence index, is omitted here and is inserted only where necessary.

To obtain the subsequent Enhancement Layer I <1,> â € ¦ I <l>, I <L> from the Base Layer (BL), obtained from the original filtered and subsampled I-frame, the frame reconstructed from the lower layers is subtracted from the frame original. The resulting difference is encoded at a given compression coefficient. It is emphasized that higher order layers achieve a better quality of the reconstructed image at the price of a lower compression efficiency. Each l-th layer I <l> is recursively encoded according to the following relations for the I-frames:

ì I 0 = JPG (F 0

I),

TO

I1 0

à = JPG [F I -IJPG (I)], (2)

<Ã ̄> I l IJPG (I l-1)) + IJPG (I l

î = JPG [F <-> 2

I - ()], l ³ 2 where it indicates a (l-1) -th JPEG decoded layer.

For interframes, the encoding process

uses the frame, ie the base layer of the intraframe, decoded JPEG, as the reference frame. Since the I-frame Base Layer is required by the encoder as a reference, at the receiver side, the I-frame Base Layer I <0> stream should be received as a minimum rate to avoid drift problems.

To encode the Base Layer P <0> of the inter-frames, after low-pass filtering and subsampling, the P-frame FP is compared to the reconstruction of the Base Layer I <0> of the encoded I-frame FIe as follows:

P <0> = JPG [IJPG (JPG (F <0>)) - IJPG (I <0>

P)] (3) The other layers at the other levels are encoded in a similar way using the lower layer as a reference: the image reconstructed from the lower layer is compared to the original P-frame FP and the current layer is recursively encoded as follows:

ìP0 = JPG [IJPG (JPG (F0

P)) -IJPG (I 0)],

TO

ÃP1 = JPG [F P-IJPG (P0) IJPG (I 0)], (4)

<Ã ̄>

îP l = JPG [F P- (IJPG (Pl-1)) + IJPG (P l <-> 2)], l ³ 2

where F <0>

P indicates the Base Layer obtained from the original P-frame Fpsubsampled and filtered.

In order to obtain a scalability with a resolution greater than that of the L layers, or layers, of the layered coding, the proposed method provides for the use of a multiple description coding technique. As can be seen from figure 2, in the scalable coding module 12 each layer I <l>, P <l> is therefore divided into a matrix of N blocks B1â € ¦BN, each j-th block Bj being independent from the others , sending this layer to a corresponding block subdivision module 13, which performs the subdivision and coding of the block. In particular, the blocks are obtained by dividing the layer into a decomposition matrix, comprising square sub-blocks adjacent to each other, as detailed in Figure 7, and of multiple dimensions compared to those of the elementary block used in the JPEG encoding, in the example 8x8 pixel. In the remainder of this description, the blocks thus obtained will therefore be referred to as descriptions. The block definition used here is the one adopted in the JPEG standard.

On the basis of user requests, for example in terms of perceived quality and available bandwidth, the selector module 14 allows the transmission of only a subset of the coded blocks.

The selection of each block is based, preferably, on the evaluation of a mean square error parameter mse in the selector module 14, a detailed diagram of which is shown in figure 3.

For the present description it is assumed that each block B has a size of MxM pixels, with M for example equal to 8 at level 0 of the Base Layer, while it indicates the jth block of the k-th frame, i.e. the block of the current frame entering the coder, specifically to selector 14. Note in this regard that each l-th layer I <l> or P <l> has its own j-th block, which therefore

should be indicated for completeness with. Each

however, block is always compared with blocks

belonging to its l-th layer I <l> or P <l>, therefore for simplicity of representation the superscript l indicating the current I <l> or P <l> layer is omitted below, leaving the superscript k which indicates to which frame in the temporal succession of frames it belongs and the subscript j which indicates which block is among the N blocks into which the layer I <l> or P <l> is divided.

Once the j-th block has been coded, the selector module 14 compares it with the j-th block of the previous frame, of index k-1, evaluating an estimator of the difference with respect to a corresponding block of a previous frame at the level of the block, ie an estimator of a quantity evaluated on the complex of the block of size MxM, for example on its pixel values. Note that each layer l has its own jth block which should be denoted by. Since

each block is compared to blocks of the same layer

, with m indicating a generic m-th frame or reconstructed image, the subscript indicating the layer l to which the j-th block belongs is omitted here for simplicity. In particular, this estimator is preferably a value of the mean square error mse of the current block according to the relation:

M M 2

mse = <1> (k) (k 1

M 2Ã ¥ Ã ¥ <X> n, m <B> j- <X -> n, m <B> j) (5) n = 1 m = 1

where it indicates a pixel at position (n, m), n and m indicating the rows and columns in the block, for the jth block of the k-th frame. For the purposes of this description for the j-th block of the previous frame, di

index k-1, from which the j-th block is taken, means a frame in its own right or, preferably as frame k-1, it is considered the last image reconstructed. If the value of the mean square error mse is greater than a given threshold THmse, which can be set, this block is transmitted. This threshold given THmse, which is representative of a minimum difference between the blocks, can be set, ie adjustable, according to user requests and the available bandwidth. The lower the given THmse threshold, the higher the quality and rate, or rate, of blocks transmitted.

On the decoding side, each block filtered by the selector module 14, ie each block, indicated below as rejected block n, which has not been transmitted, is replaced by the corresponding block of the previous frame. Since the substitution of the blocks with error value mse below the given threshold THm can lead to the propagation of a permanent distortion in the video sequence, in order to reduce this effect, in addition to the evaluation of the value of the mean square error mse, which in various embodiments could be used alone, it is envisaged to introduce an aging counter parameter, or aging, in transmission aj (or aj <k>), wanting to address the frame) for the j-th block and a threshold of Ath aging count, tx. The aging counter parameter aj of each block is incremented each time the block Bj is discarded, and reset, i.e. reset to zero, when the block is transmitted on a channel 15. Therefore this aging counter parameter aj substantially indicates the number of transmission attempts of the block during the transmission of the sequence S. If the aging counter parameter aj reaches the aging count threshold Ath, tx, the j-th current block is sent in any case, regardless of the given threshold THmse, and the counter aging is reset. The current block is then transmitted if one of the following two conditions is satisfied:

ìmse> TH mse

TO

îaj ³ A (6)

th, tx

that is, if the mean square error mse is greater than the given threshold or if the aging counter parameter aj is greater than the aging count threshold Ath, tx.

Figure 3 shows the detail of the selector module 14, which receives the j-th current block as input, and comprises, in parallel, a first buffer module 141, i.e. a buffer, to temporarily store this block and a second buffer module 142 preceded by a delay line 143 for storing the preceding block. The outputs of said first module 141 and said second module 142 are sent to a calculation module 144 which implements the calculation of the error mse as indicated in equation (5). In a comparison module 145 it is therefore evaluated whether the error mse is greater than the given threshold THmse, as in the first of the two conditions (6). In the positive case, in an aging reorganization module 146 the aging parameter aj is reset, i.e. set to zero and the j-th current block is transmitted on channel 15. In the negative case, the control passes to a reject module 147 which provides for the verification of the aging parameter and possibly discarding this j-th current block. For greater clarity of description, Figure 3 shows this discarded block n, although it is generally simply discarded by failing to transmit it. This reject module 147 also includes an aging comparison module 148, which receives the aging counter parameter aj, which is calculated, in particular increased, at module 141 upon receipt of the j-th current block, and evaluates whether this aging counter parameter a <k>

j is greater than the Ath aging count threshold, txin transmission. If so, in the aging reorganization module 146 the aging parameter aj is reset, i.e. set to zero, and the j-th current block is transmitted. If not, in an aging increment module 149 the aging counter parameter aj is incremented and the j-th block B <k>

j is omitted, that is, it is not transmitted and discarded, this being symbolized in figure 3 by the discarded block.

The process and system according to the invention in the application to channels with packet transmission is now described, illustrating the process of packet loss concealment.

In an ideal transmission, without packet loss, the decoding process is closely related to the encoder: the decoder is performed an inverse series of operations compared to those performed by the encoder. For example, indicating with the l-th layer, for an inter-frame, reconstructed by the decoder, it is evaluated as the JPEG decoding of the difference between the l-th layer and the previous layers, plus the JPEG decoding of the previous layers. Thus, the layer is evaluated as follows:

I <Ë † l> = IJPG (I <l>) + IJPG (I <(l-1)>) - IJPG (I <(l -2)>), l ³ 2 (7) This decoding process can be extended to any layer block.

In decoding, the blocks are juxtaposed according to the order used in the decomposition matrix used for the subdivision operation.

A respective reception aging counter parameter aj, rx and a reception aging threshold Ath, rx are also introduced in the decoding process. Whenever the j-th block is lost, i.e. it is left out to the selector module 14 in transmission or transmitted but not correctly received (for example due to an error along the transmission channel), the reception aging counter aj, rx is incremented and the block is replaced by the previous block of the l-th layer.

When the reception aging counter parameter aj, rx reaches the reception aging threshold Ath, rx, the missing block is no longer replaced, it produces a "hole" in the l-th layer. The reception aging counter parameter aj, rx is reset upon each successful reception of the block.

Below is a possible procedure for setting the data mse threshold and optimizing the aging parameters. On the basis of the boundary conditions imposed by the access network and user requests, a given bit-rate R is considered for a wireless channel with a given rate of packet loss Plr. Hence, the average bits for each transmitted frame are given by the relation:

b <- 1>

<pf> = R [bps] ́ (Rf [fps]) (8) where Rfà is the frame-rate, usually set at 25-30 fps. Assuming a packet length of Spbits, the average number of packets per frame is

à © ù

<P> pf = <b> pf

ê ú

êê S (9) p úú

On the basis of the boundary conditions and the channel parameters imposed by the system, we can therefore express the minimization of the expected distortion E [D] for a video sequence, evaluated as the distortion of each single frame averaged over the entire sequence

P pf

E [D] = Ã ¥ D <(> Bk <)> Pk (10) k = 0

where Bkra represents the number of information bits correctly received when the first k transmitted packets are correctly received, and D (Bk) denotes the distortion, preferably calculated as mse. Pkra represents the probability of losing the k-th packet, and depends on the packet loss Plr.

Therefore, the optimization problem to solve is

{Th, mAin, A} = {E [D] (Plr, bpf, Sp)} (11) mse th, tx th, rx

which allows to optimize the expected distortion E [D], identifying the threshold parameters given mse and aging thresholds as a function of the values of as a function of packet loss rate Plr, number of average bits per frame and packet length Sp of a certain network where the video sequence S is being broadcast. Note that the mse data threshold and the aging parameters allow to maximize system performance and optimize the quality of the service even when users perform a handoff, i.e. the transfer of a data or voice session, from a high-speed technology to lower bit-rate technology.

Figure 4 shows a detailed diagram of an embodiment of the layered coding module 11, indicated with 41, which implements a layered coding procedure described previously with reference to equations (1), (2), (3 ), (4). This layered coding module 11 operates in this case with L = 3, producing an output for the intra-frame therefore a Base Layer I <0> and two Enhancements

Layer I <1> and I <2>. Therefore, this structure has three respective branches t0, t1, t2, which receive the I-frame FI as input. The inter-frame is sent to a similar structure, with three respective branches tâ € ™ 0, tâ € ™ 1, tâ € ™ 2 which determine a Base Layer P <0> and two Enhancement Layer P <1> and P <2>. In figure 4, as well as in the following figure 5, identical references indicate modules that perform substantially equivalent functions, even if they may have implementation differences linked to the quantities at their inputs.

The I-frame FI is therefore received on the first branch t0, which is filtered in a low-pass filter 41 and under-scaled in a downscaling module 42, giving rise to the filtered and subsampled I-frame, which is encoded with JPEG encoding with a JPEG encoder of the Base Layer 430, ie an encoder that produces the DCT coefficients of the layer at level 0. In general, the coefficients used for JPEG compression of different levels are in fact usually chosen with different values.

This Base Layer I <0> is produced at the output of the first branch t0, but it is also sent on the second branch t1 to a JPEG 44 decoder, whose decoded output is sent to an interpolator block 45 and then to a subtractor block 46 which makes the difference with the original I-frame FI. This difference is sent to a level 1 431 encoder, which produces the DCT coefficients of the layer at level 1, I <1>, i.e. the first Enhancement Layer, which is output on the second branch t1. The first Enhancement Layer I <1> is also sent to the third branch t2 through a further JPEG decoder 44 of the third branch t2, whose output is summed in an adder 47 of the branch t2 with the decoded version of the Base Layer I <0 > coming from the first branch t0. The resulting sum is then subtracted from the original I-frame FI, in a subtractor block 46 of the t2 branch, which outputs, after JPEG encoding in a 2432 level encoder, the second Enhancement Layer I <2>.

The structure of the decoder module 41 which operates on the P-frame, as mentioned, is similar, with three branches tâ € ™ 0, tâ € ™ 1, tâ € ™ 2 receiving this original P-frame in parallel, the P-frame sent on the first leg tâ € ™ 0, it is also subjected to a filtering in a low pass filter 41 and to an underscale operation in a downscaling module 42. The P-frame underscale is filtered however, compared in the case of the I-frame, before the JPEG decoding on the first branch tâ € ™ 0 in a decoder 44 the JPEG is recoded with a level 0 encoder 430 of the branch tâ € ™ 0. Then the decoded Base Layer I <0> from the corresponding JPEG 430 encoder on the first branch t0 of the I-frame encoding is subtracted from the decoded and recoded version of the frame, in a subtractor module 46 of the first branch tâ € ™ 0 to which the decoded Base Layer of the I-frame taken from the corresponding JPEG 430 encoder in the coding portion of the I-frame is also sent. The difference in output from the subtractor module 46 of the first branch tâ € ™ 0 is JPEG encoded by a further JPEG 430 encoder and is sent to the output of the first branch tâ € ™ 0 as the Base Layer of the P-frame, P <0>, as well as © sent to the second branch tâ € ™ 1 through a decoder 44, whose output is sent to an adder 47 of the second branch tâ € ™ 1 to be added again to the decoded Base Layer I <0>, and, after interpolation in a interpolator block 45 of the second branch tâ € ™ 1, the result of this sum is subtracted in a subtractor module 46 of the second branch tâ € ™ 1 from the original frame. The difference is JPEG encoded in a level 1 encoder 431e started at the output of the second branch tâ € ™ 1 as the first Enhancement Layer, P <1>. This first Enhancement

Layer, P <1>, also decoded in a decoder 44 of the third branch tâ € ™ 2 and added with the interpolator output of the first branch tâ € ™ 0 in an adder 47 of the third branch tâ € ™ 2, whose resulting sum is subtracted from the original P-frame to obtain the second Enhancement Layer, P <2>, output on the third branch tâ € ™ 2.

Figure 5 schematically represents an embodiment of decoder 51 which implements equation (7). It is simpler than encoder 41 and includes a JPEG decoder 54 in parallel for each layer I <0>, I <1>, I <2>, P <0>, P <1>, P <2>, in entrance. The

Base Layer I <0> is then sent to an interpolator 55 and then output as IQ0 frame, i.e. with QoP (Quality of Picture) 0. The first Enhancement Layer I <1> output from the respective decoder 54 is sent to an adder 57 which adds it to frame Q0 to create a frame IQ1 with QoP 1. The second Enhancement Layer I <2> output from the respective JPEG decoder 54 is sent to another adder 57 which adds it to frame IQ1 with QoP equal to 1 to make an IQ2 frame with QoP2.

As regards the decoding portion of the P-frame, downstream of the JPEG 54 decoder which decodes the Base Layer P <0>, an adder is provided to add the Base

Layer P <0> decoded with the Base Layer I <0> decoded and send the sum to an interpolator 55. The rest of the structure downstream of the interpolator 55 is similar to that described for the decoding of the I-frames and determines three P-frames PQ0, PQ1 and PQ2, with QoP 0, 1, 2 respectively. As shown in the figure for both the I-frame and the P-Frame it is then possible to select the frame with the desired QoP.

Figure 6 provides by way of example a comparative diagram of the performances of a process according to MPEG4-FGS (Fine Granularity Scalable video coding) and the process according to the invention.

For video streams there is no objective technique for assessing the quality perceived by the user. The benchmark that has long been used to assess video quality is the ITU-R BT.500 recommendation, while more recently the ITU-T organization developed the P.910 recommendation for multimedia quality assessment and the ITU-T Study Group 9 is working in conjunction with the Quality Video Expert Group (VQEG) on maintaining and improving recommendations P.910, P.911, P.920, P.930, P.931, J.146, J. 148, J.149. The objective reference model proposed by the VQEG is the Peak Signal to Noise Ratio (PSNR). This PSNR metric is used in the comparison of figure 6 between the proposed codec and the state of the art MPEG-4 FGS. The FGS encoding is adopted by the MPEG-4 standard for the efficient and flexible distribution of multimedia streams over heterogeneous networks, ie networks comprising different wired and wireless physical media.

Figure 6, which shows on the horizontal axis values of probability of packet loss Plre on the vertical axis values of PSNR, therefore shows the curve of the FGS standard (MPEG4-FGS) (curve point point) compared to different curves for different threshold values given THmse, in particular zero value (continuous curve) as well as further four curves for respective pairs of parameter values: THmse = 10 Ath = 1; THmse = 10 Ath = 4; THmse = 30 Ath = 1; THmse = 30 Ath = 4, as indicated in the legend of figure 6. These curves are therefore obtained with the proposed coding procedure, and represent a fine adjustment of the coding parameters. In any case shown, with a probability of packet loss greater than zero, the encoding / decoding process and encoder / decoder apparatus according to the invention have completely higher PNSR values than the coding of the FGS standard.

Figure 7 is a schematic representation of an image frame of the sequence S of image signals encoded by the method according to the invention, for example an intra-frame FI, and of the subdivision into blocks. The larger squares indicate the Base Layer blocks discarded by selector 14 in figure 3, indicated with n (I <0>), while the smaller squares inside them indicate the Enhancement Layer blocks discarded by selector 14 in figure 3, denoted by n (I <l>). The areas

empty, which are occupied by similar blocks (I <0>) (I <l>) not highlighted in figure 7, are those relating to the blocks not discarded, that is, transmitted. HM indicates an area characterized by high movement, and, as you can see, in this area there is the greatest number of blocks transmitted. Ultimately, the areas with high movement are those that vary more rapidly and therefore in the comparison module 145 which evaluates whether the error mse is greater than the threshold given THm give a positive result and are therefore more likely transmitted on channel 15.

Therefore, the advantages of the coding process just described are clear.

The method according to the invention advantageously provides for the use of a layered encoding, which is in itself less efficient from the point of view of compression, however the encoding with multiple descriptions makes the transmission more robust and, allowing not to transmit all the blocks, it allows to recover compression efficiency.

The method according to the invention allows, by means of scalable multi-layer coding operations, to meet the multi-user requests in a heterogeneous network. On the basis of access network technology and user requests, the best threshold values can be adjusted, specifically the mse threshold and aging parameters, which allow to obtain an encoded video streaming capable of adapting the constraints imposed in terms of quality and bandwidth.

In particular, the processes and apparatuses according to the invention are particularly suitable for solving scalability problems for video streaming on wireless networks. The coexistence of multiple access technologies determines that the fourth generation wireless systems under development are characterized by increasing heterogeneity. In this scenario, one of the purposes is to provide a so-called â € œalways best connectedâ € (ABC) service and in this the type of access network plays a primary role. The ABC service is enhanced by the optimization of the source coding, according to the invention.

From a single coding process, the process according to the invention is advantageously able to meet multi-user requests with the best quality-coding trade off. Beyond scalability, the proposed encoder is characterized by robustness to packet loss. When a non-zero packet loss is considered, the proposed procedure achieves a quality even 7 dB higher than the MPEG-4 FGS codec.

Naturally, the principle of the invention remaining the same, the details of construction and the embodiments may vary widely with respect to those described and illustrated purely by way of example, without thereby departing from the scope of the present invention.

Claims (15)

CLAIMS 1. A method of coding the source of image signals of a scalable type, in particular for transmission on channels with packet transmission of telecommunications networks, which comprises operating on a sequence (S) of digital images organized in frames (FI, FP , I, P) a layered coding to obtain a plurality of layers (I <l>, P <l>), including a Base Layer (I <0>, P <0>) which is associated with a first level of image quality and one or more Enhancement Layers (I <l>, P <l>) which are associated with respective higher levels of image quality, characterized by the fact that said encoding procedure further comprises encoding the multiple descriptions (12) of each of said layers (I <l>, P <l>), said coding to multiple descriptions (12) including the operations of: divide (13) each of said layers (I <l>, P <l>) into a respective matrix of blocks (B1â € ¦BN), each independent of each other, in said matrix of blocks (B1â € ¦BN) of the layer (I <l>, P <l>) select (14) blocks for transmission through the steps of: calculate (144) for each current block () of said layer (I <l>, P <l>), at the block level, a difference estimator (mse) with respect to a corresponding block of a previous frame (), and compare (145) said estimator (mse) with a settable threshold (THmse) of difference to evaluate whether to transmit the current block (). 2. Process according to claim 1, characterized in that the operation of calculating (144) said difference estimator (mse) comprises evaluating a mean square error between pixel values () of said current block () and of the corresponding block () of a previous frame. Method according to claim 1 or 2, characterized in that it further comprises calculating (149) for each current block () an aging value (aj) which is incremented each time the current block () is discarded for transmission following said comparison operation (145), and reset (146) when the block is transmitted, and to compare (148) said aging value (aj) with an aging threshold (Ath, tx), transmitting the block anyway current () if the corresponding aging value (aj) exceeds said aging threshold (Ath, tx). 4. Process according to claim 1 or 2 or 3, characterized in that said layered coding operation (11) comprises obtaining a Base Layer (I <0>) from an intra-frame (FI) in said sequence (S) by carrying out an encoding with compression of said intra-frame (FI) after a low-pass filtering and a downsampling and to obtain at least one Enhancement Layer (I <l>) by subtracting one or more frames reconstructed from lower layers from the original intra-frame (FI) , encoding the resulting difference at a given resolution and a given compression coefficient. 5. Process according to claim 4, characterized in that said layered coding comprises obtaining a Base Layer (P <0>) from an inter-frame (FP) in said sequence (S) by compressing the difference between said inter-frame frame (FP), first compressed and then decompressed, and of said Base Layer (I <0>), decompressed, of the inter-frame (FP). 6. Process according to claim 4 or 5, characterized in that said sequence (S) comprises an intra-frame (FI) and one or more inter-frames (FP) referred to said intra-frame (FI). 7. Process according to one or more of claims 4 to 6, characterized in that said compression coding operation is a JPEG coding. 8. Process for decoding coded image signals according to the coding process according to one or more of claims 1 to 7, characterized in that it comprises operating a corresponding layered decoding to obtain from said Base Layer (I <0> , P <0>) and Enhancement Layer (I <l>, P <l>) transmitted a plurality of reconstructed frames (IQO, IQ1, IQ2, PQO, PQ1, PQ2) which are associated with different levels of image quality . 9. Decoding method according to claim 8, characterized in that it comprises calculating a respective reception aging counter parameter (aj, rx) and setting a reception aging threshold (Ath, rx), increasing each time the current block () is omitted in transmission, or in any case not correctly received, said reception aging counter (aj, rx) and replace the current block () with the previous block (), said replacement operation being performed only until the receive aging counter parameter (aj, rx) reaches the receive aging threshold (Ath, rx). 10. Process for coding image signals according to one or more of claims 1 to 8 and decoding according to one or more of claims 8 to 9, characterized in that it comprises regulating said settable threshold (THmse) and said aging thresholds (Ath, tx, Ath, rx) for transmission and reception by finding the corresponding values that minimize an expected distortion (E [D]) of the sequence (S) of digital images, with respect to given values of packet loss rate ( Plr), average bits per frame () and packet length (Sp) characteristic of a communication network in which the sequence (S) of digital images is transmitted. 11. Apparatus for coding image signals in a sequence (S) of digital images, in particular for transmission on channels with packet transmission of telecommunications networks, characterized in that it is configured to implement the coding process according to one or more than claims 1 to 7. 12. Coding apparatus according to claim 11, characterized in that it comprises a module (11, 41) configured to perform said layered coding operations and a module (12) configured to perform said coding operation on the multiple descriptions of each of said layer, which includes modules to divide (13) each of said layers (I <l>, P <l>) into a matrix of blocks (B1â € ¦BN), each independent of each other and modules ( 14) to operate in said block matrix of the layer (I <l>, P <l>) a selection of blocks for transmission. 13. Coding apparatus according to claim 11 or 12, characterized in that said modules (14) to operate in said block matrix of the layer (I <l>, P <l>) a selection of blocks for transmission are also configured to calculate an aging value (aj) for each current block () and compare said aging value (aj) with an aging threshold (Ath, tx), still transmitting the current block () if the corresponding aging value ( aj) exceeds said aging threshold (Ath, tx). 14. Decoding apparatus configured to implement the decoding method according to claim 8 or 9 characterized in that it comprises a module configured to calculate a respective reception aging counter parameter (aj, rx) and set a reception aging threshold (Ath, rx), increment each time the current block () is omitted in transmission, or in any case not correctly received, called reception aging counter (aj, rx) and replace the current block () with the previous block () , said replacement operation being performed only until the reception aging counter parameter (aj, rx) reaches the reception aging threshold (Ath, rx). 15. Computer product which can be directly loaded into the memory of a digital computer and comprising portions of software code for carrying out the method according to any one of claims 1 to 9 when the product is executed on a computer. All substantially as described and illustrated and for the specified purposes.