System, packet structuring and device for processing output information from a signal encoder.

0001The present invention refers to a system for structurizing the information outgoing from an encoder of signals, specially video signals and for organizing the generated information in information units, called cells or packets suitable for being transmitted and treated by new generations of transmission networks. e.g. ATM networks (Asynchronous Transfer Mode). The invention also concerns the devices used for the application of the considered system.

0002It is well-known the encoding of video signals uses compression technics: the so compacted signals are generally transmitted on the present transmission networks at a bit rate substantially constant. Since the encoders generate data streams at variable bit rate, it is necessary the introduction of buffer memories or buffers, that causes disadvantageous variation of the signal quality.

0003Recently transmission networks able to treat information streams at variable bit rate. e.g. ATM networks have been proposed. However these networks are characterized by a series of problems, such as the probability of losing packets and/or delivering some of them with a wrong content: in addition the packet loss can greatly increase in the so called congestion situations.

0004Aim of the present invention is the utilization of an encoder type and a method to organize the video information with the end of optimizing the whole system as regards the channel errors and cell losses caused by the so called networks, in order to obtain a simple and efficient management of the lost and/or wrongly received cells and the following error concealment.

0005The use of the invention verifies the presuppositions that make possible the error concealment, respectively: the localization of the coded and packetized signals. According to the stratification of the functions proposed as methodology for the ATM network, the detection and the following error concealment are included in the high levels with typical functions of the application and ATM adaptation layer (AAL).

Basic structure of the encoder

0006According to what above said and particularly with the localization of the coded signal, the compression technics that can be used for the redundancy reduction of the television signal must guarantee a coding carried out independently in each field in parts of images named blocks, so that the loss of some blocks does not influence the decoding of the remaining blocks which are received correctly. According to a first feature of the invention the encoders used in the system, in conformity with the invention are preferably selected from the group of encoders utilizing adaptive prediction schemes typically intrafield, interpolated intrafield and interframe, with motion compensation or no motion compensation, all based on the segmentation of the image in blocks. It is well-known to consider as very efficient for the television signal coding the prediction of interpolated intrafield and interframe type; the motion compensation using the "block matching" is generally added to improve the efficiency.

0007For changes of scene or in any case for fast motions it is preferably used an intrafield coding, that can also be imposed periodically following a policy of "refresh", to avoid the accumulation of errors that are necessarily caused by the prediction. Moreover it is well-known that the hybrid coding schemes combine the temporal prediction with domain transform operations, reducing the spatial correlation in addition to the temporal correlation: it is normally uilized the bidimensional cosine transform, that allows to use fast algorithms and implicitly leads to segmentation of the image in blocks.

Transmission

0008With the aim of better understanding the problem, we can refer to the structure in transmission of a conventional hybrid encoder (see e.g. S. Cucchi, F. Molo, "DCT Based Television Codec for DS3 Digital Transmission", SMPTE, October 1988) as indicated in Figure 1a. Now we anlyze the scheme from left to right side: <ul id="ul0001" list-style="none"><li>1a) the input blocks S1 belong to the field under analysis. In all the algorithms based on adaptive prediction schemes it is searched the block belonging to the interpolated preceding field (S2) or to the preceding frame (S6), that gives the better prediction of the block under test. Among the various possible selections, it is preferred the one that supplies a prediction error (S3) equal to the difference between the input block (S1) and the estimated block (S2), with a minimum energy content. The comparison criteria among the various possible selections are based on energy measures and crossed comparisons or with respect to thresholds; Figure 1a indicates with B7 the comparison device that receives as input the block under analysis S1, the possible prediction blocks S5, S6, S7 coming from B9, B10 and B11 (estimation devices of the moving for the motion compensation). The aim is the transmission of the prediction error with the minimum information content (S3), this implies for a correct reconstruction in reception, to transmit the information about the selected predictor (S8). In the case of prediction with motion compensation the choosen motion vectors (S10) are also included, in addition the information of refresh is sent, if it is foreseen (S4 in input at B7 and S9 in output from B7).</li><li>2a) The selected prediction error (S3) is submitted in B1 to a domain transform (DCT = Discrete Cosine Transform), followed eventually by a scaling (B2), that weights the psychovisual contibution of the various transformed coefficients.</li><li>3a) Different types of quantizers (B2) can be selected in adapative manner, in order to guarantee a signal to noise ratio, which is almost constant on the average and decided by the quality control (B3). The quantization also depends on the state of the encoder-ATM network interface buffer memory (B15) through a feedback path (S11). This buffer memory allows a control of the generated bit rate reducing the quality level, in order to adjust the behavious of the COD encoder conformably to the statistical parameters negotiated with the ATM network during the call set-up phase. The type of used quantizer, possibly in common to many blocks, is an information that must be transmitted (S12).</li><li>4a) In transmission and congruously in reception the quantized prediction error (S13) is rescaled (B5) in manner dual to an eventual sclaing (B2), antitransformed (B6) and added to the prediction block (S2). Obviously the reconstructed block (S14) differs from the original block (S1), since the quantization is unreversible.</li><li>5a) The quantization of the prediction errors is followed by a combination of encoding utilizing runlegnth of zero's and words with variable length. The determination of symbols to be encoded is based on scannings of zig-zag type or similar ones (B4). Finally each error prediction block supplies a different information quantity (S15), that is a stream at variable bit rate. If the motion vectors (S10) indicated previously are encoded by codes with variable length words, also this information is generated at the variable bit rate.</li></ul>

Organization of the video information

Classification of the information

0009The aim of the video information structure, as reported in the following, is to packet the blocks and the information associated to these ones in one or more cells, in order to limit and locate the influence of the cell loss.

0010The video information is organized in VIDEO UNITS: according to a feature of the invention each UNIT corresponds to a space location of prefixed dimension of the television image, i.e. having a fixed number of blocks, and it is divided onto a variable number of cells. For a colour television image, the luminance and chrominance, related to a same space part and generally processed in a different manner, are transmitted as a single VIDEO UNIT.

0011The transmission order of the luminance and chrominance inside the same VIDEO UNIT guarantees the separation of the components, without further added information. However it is possible to process and transmit the luminance and chrominance on different VIDEO UNITS. The selection of the part of television image associated to the VIDEO UNIT generally depends on the type of coding system that is used and includes one or more blocks in which the same image was divided.

0012A particularly advantageous and therefore preferred fulfilment form of the invention foresees that for each VIDEO UNIT, the information to be transmitted is divided in three different classes: a) normal, B) privileged, C) auxiliary. <ul id="ul0002" list-style="none"><li>A) The normal class A) includes the prediction errors. The information outgoing from the encoder for all blocks belonging to a VIDEO UNIT is inserted in consecutive cells. As the data stream generated is made up with variable bit rate, the number of cells differs among the VIDEO UNITS: this number is considered an information belonging to the auxiliary class C).</li><li>B) The privileged class B) includes the selections made in all the blocks belonging to a VIDEO UNIT, comprising the motion vectors in the case the prediction with motion compensation is implemented together with the used quantizer type. This kind of information can be common to all blocks forming the VIDEO UNIT or differs among the component blocks. The said privileged class B) also includes the information common to many VIDEO UNITS and distributed among the same UNITS according to a suitable assignment policy. For example, if the VIDEO UNIT corresponds to a stripe, the above mentioned information may be common to a field or frame or many frames, as the specifications of the used television standard. The data stream generated by the information of this class B) can now be divided in a part at constant bit rate, typically for the selected predictors and the type of quantizer and a part at variable bit rate, generally coming from the coding with variable word length of the motion vectors. Therefore the number of cells related to the constant bit rate can be fixed a priori, while the number related to the variable bit rate is assumed as information belonging to the auxiliary class C).</li><li>C) The auxiliary class C) includes all the information that allows a simple and efficient management of the VIDEO UNITs in terms of the detection of lost or wrongly received cells and the following error concealment For example the number of cells generated by the variable bit rate data stream belonging to the normal A) and privileged class B) were formerly indicated as forming a part of the class C).</li></ul>

0013According to another feature of the invention the remaining information depends on the protection method used for the information belonging to the various classes and on the strategy of separation for the consecutive VIDEO UNITS.

Video packet

0014A VIDEO PACKET formed by two parts: E) HEADER, F) BODY is associated to each VIDEO UNIT. The header E) and the body F) differ both for the contained information classes, respectively privileged B) and auxiliary C) information in the header E) and normal information in the BODY F) and for the different method of protection applied.

0015Figure 2 indicates the structure of the PACKET according to the invention, in this particular case the VIDEO PACKET is referred to the cells with 48 byte of information field (payload) available with a direct access at ATM layer.

0016E) THE HEADER E) precedes the BODY F) in the structure of the VIDEO PACKET and includes: <ul id="ul0003" list-style="none"><li>a) an identifier (VUI = Video Unit Identifier: e.g. 1 byte if it is assumed as order number), which allows the distinction among following VIDEO PACKETS, that is, owing to the established correspondence, among VIDEO UNITS trasmitted in succession.</li><li>b) An alignment word (HAW = HEADER ALIGNMENT WORD: e.g. two byte), that allows to identify the start of the VIDEO PACKET. The contemporaneous use of a similar alignment word (BAW see a')), allows to distinguish inside the same PACKET the HEADER E) from the BODY F).</li><li>c) The information of the privileged class B) at constant bit rate (CBRH = Constant Bit Rate HEADER) typically the selected predictors and the type of quantizer, related to the present VIDEO UNIT and to other two VIDEO UNITS located in the original image at a suitable spatial distance. In a fulfilment form of the invention, when the VIDEO UNIT is caused to correspond to a stripe, the three VIDEO UNITS are taken at a distance of half field. For example using a format 4:2:2, images of 720*288 pixel per field, blocks of 8*8 samples, a stripe generates 90 luminance blocks and 90 chrominance blocks for the two colour components. Then, e.g. a byte can be associated to the selected predictions related to each block belonging to a quadblock y/cb/y/cr, corresponding to a same video part of 16*8 pixel. Figure 2 indicates the information for the whole stripe (SC= Strip Choice, here e.g. 45 byte, FPSC= First Previous SC, SPSC= Second previous SC). Two different identifiers relative to the luminance and chrominance components are reserved for the type of quantizer and refresh, e.g. this information could be in common to a whole stripe (LQ/R = Luminance Quantization/Refresh: e.g 1 byte, CQ/R = Chrominance Q/R, FPLR/R = First Previous LQ/R, FPCQ/R, SPLQ/R = Second Previous LQ/R, SPCQ/R). According to a further feature of the invention the protection on the information of privileged class is carried out by repetition: in reception the information repeated three times is cumulated with a delay that is correlated to the preset space distance, then a decision at majority byte by byte is applied. The rule also holds if the delay is imposed at the transmission side.</li><li>d) The information of privileged class B) at variable bit rate (VBRH = Variable Bit Rate HEADER), typically the motion vectors, following the same phylosophy of protection by repetition of the point c). This information either can be related to the considered VIDEO UNIT and other two UNITS or it represents a part of an information in common to many VIDEO UNITS, that being repeated three times, is distributed among the various UNITS according to a suitable policy. This method also holds if the motion vectors are coded with fixed length words, obviously in this case the number of cells can be fixed a priori and not considered as auxiliary information C). Moreover this rule may be also extended to the information with fixed or variable bit rate in common to many UNITS, typically video standard specifications. (VF = Video Format: e.g. 1 byte, FPVF = First Previous VF, SPVF = Second Previous VF).</li><li>e) The information of the auxiliary class C) equal to the cell number: <ul id="ul0004" list-style="dash"><li>forming the BODY F) (BNC = Body Number Cell: e.g. 2 byte, FPBNC = First previous BNC. SPBNC = Second Previous BNC) and</li><li>comprising the information of HEADER E) at variable bit rate, provided that it exists as indicated at point c) (HNC = Header Number Cell; e.g. 1 byte, FPHNC = First Previous HNC, SPHNC = Second Previous HNC).</li></ul> The protection applies the same rule used at point c), as this information is associated to each VIDEO UNIT.</li><li>f) The information belonging to the auxiliary class C) and not indicated formerly, that, according to another advantageous feature of the invention, develops a protection function onto the BODY F) content. A maximum fixed number of CRC (Cyclic Redundancy Check) protects the prediction errors included in the BODY F), in order both to allow the localization of the lost cells and to guarantee that the received cells are exempt from errors due to the transmission channel. The kind of feature can be implemented utilizing at the same time the CRC codes, the previous information BNC belonging to the same auxiliary class C) and a window method. Since ther BODY F) is composed of a variable of cells, the efficiency of the CRC codes, in terms of CRC number really utilized, may be different among the VIDEO UNITS. The CRC number depends on the locking characteristics of the system, based on the recognition of end of block symbols generated as pseudorandom sequences by suitable devices (scrambler sistems). This system is preferably the system described in a former application of patent in the name of the Applicant. In transmission the cells including the prediction errors related to the VIDEO UNIT are scanned in succession until it is cumulated the necessary number of words that the selected code can cover or until the end of the BODY F). In reception, if errors are detected, all the part of VIDEO UNIT covered by the CRC code, generally extended onto a no integer number of cells, is deemed lost, as if all the cells of this portion were not received. As the CRC are related to the prediction errors of a VIDEO UNIT, the protection follows the same rule used in c) (CRC: e.g 3 byte, FPCRC = First Previous CRC, SPCRC = Second Previous CRC).</li><li>g) An additional information (VSR = Video Sync. Recover, e.g. 3 byte) that allows the recovery in reception of the video synchronism.</li><li>h) A residual part (SPARE) for further specifcications and/or the case in which the information field size of the cell is less than the one intially foreseen.</li></ul>

0017F) The BODY F) follows the HEADER E) in the structure of the VIDEO PACKET and includes: <ul id="ul0005" list-style="none"><li>a') an alignment word (BAW = Body Alignment Word; e.g. 2 byte) as already pointed out in b).</li><li>b') The prediction errors related to the blocks forming the VIDEO UNIT inserted on consecutive cells, so as they are supplied by the encoder. The cell number depends on the selected quantizer and in any case it changes among the VIDEO PACKETS (VBRB = Variable Bit Rate Body). The last cell is generally filled with dummy bits in order to avoid the propagation of a VIDEO UNIT towards contiguous VIDEO PACKETS.</li></ul>

0018Figure 1b shows the construction of the VIDEO PACKET following the above mentioned specification. We shall always examine the scheme from left to right side: <ul id="ul0006" list-style="none"><li>1b) a processor (B14) receives the inputs, related to the privileged information B) (S8, S9, S10, S12, S16) and normal information A) (S15) and it adds the auxiliary information C) generated internally in order to obtain the prefixed VIDEO PACKET format. Tthe protection method used also requires a VIDEO PAKET memory (in B14), to build the whole information associated to a VIDEO UNIT, e.g. a stripe, a field memory (B13) for the information included in the HEADER E) and a VIDEO UNIT memory for the BODY F) (B12). The processor (B14) accumulates in B12 the prediction errors of a whole VIDEO UNIT, it carries out the calculation of CRC codes and generates the HEADER E) of the PACKET related to the VIDEO UNIT under test, using the content of B13, which corresponds to the privileged B) and auxiliary C) information of other two VIDEO UNITS located at a spatial distance respectively of half field and of a whole field. To allow the repetition of the described process the information B) and C) of the considered VIDEO UNIT is stored in B13 in the corresponding location.</li><li>2b) The cells forming a VIDEO PACKET are delivered to the network through the buffer memory (B15) following a suitable strategy. As above stated, the presence of the said buffer memory (B15) allows to adjust the bit rate generated by the encoder COD conformably to the statistical parameter negotiated with the network during the call set up phase (S11). This adjustment may be forced by the presence of a congestion signal coming from the network to the encoder (S17), if this signal is foreseen.</li></ul>

Error concealment

0019According to a further feature of the invention, the error concealment is founded on the assumption of replacing in reception the reconstructed block with prediction block, if the prediction error is not received or it is deemed wrong.

0020The prediction block, selected in transmission and belonging to the prevoius interpolated field or to the previous frame, is the block that better estimates the block under test or equivalently supplies a prediction error with a minimum energy content.

0021In case of unforced intrafield selection, that is not coming from the refresh, the prediction block does not exist becouse the prediction error coincides with the block under test, therefore it is assumed to take from the frame memory in reception the nearest homologous block, that is belonging to the preceding interpolated field. However it is also allowed to take the block in the same spatial position but belonging to the preceding frame.

0022In case of forced intrafield selection, that is coming from the refresh, if the selected predictors used in absence of refresh are always transmitted together with the refresh information (1 bit for block or less, as generally the refresh is forced on image parts), it is possible to fall down into the general procedure of replacement with the prediction block, that supplies the better estimate. This concealment method requires that the selections together with the motion vectors, in case of prediction with motion compensation, are received correctly for each block belonging to a VIDEO UNIT. This is guaranteed by the repetition of this information for three times and by the majority decision.

Reception

0023As in transmission we can refer to the structure in reception of an hybrid encoder with error concealment, that uses the format of the established VIDEO PACKET, as indicated in Figure 3b.

0024We analyse the scheme from left to right side: <ul id="ul0007" list-style="none"><li>1c) a receiving buffer (B1') collects the cells, intended as the only information field, delivered to the ATM network.</li><li>2c) The analysis made by the device B2' on the alignment words (HAW and BAW) and on the identifier of VIDEO UNIT (VUI), included in univocal manner into the VIDEO PACKETS, allows to distinguish the contiguous PACKETS and inside the same PACKET the HEADER E) and BODY F). The VIDEO PACKETS received and identified are stored in different locations, respectively for the HEADER E) and the BODY F). For example, if the VIDEO UNIT corresponds to a stripe, the repetition of the privileged information B) and auxiliary information C) in the HEADER E) of three diffent PACKETS interlaced on the distance of a field, requires at least a local memory and a field memory for the HEADER E). If e.g. the device B2' is charged with the task of distinguishing the information B) and C) related to three VIDEO UNITS included in the HEADER E) of the VIDEO PACKET under test it is possible to make easier the comparison operations. In fact this classified information may be stored in three field memories (B3'), each representing the number of repetitions, in the locations corresponding to the associated UNITS. On the contrary the prediction errors included in the BODY F) are accumulated in a memory (B4') of at least a field, the circularly management of this memory by means of tables of pointers (B5') associated to the VIDEO UNITS make possible the distinction among BODY F) belonging to contiguous VIDEO PACKETS. The loss of cells comprising the alignment words (HAW and BAW) or the identifier of VIDEO UNIT (VU) allows a delayed identification and an incomplete storage of the VIDEO PACKET.</li><li>3c) The privileged B) and auxiliary C) information associated to a VIDEO UNIT can be obtained after a delay at least equal to the storage of the three repetitions distributed in the HEADERS E) of the VIDEO PACKETS, using a majority decision byte by byte (in B6') on the above mentioned information content repeated for three times. For example, using a VIDEO UNIT equal to a stripe and a delay of a field it is possible, for each identified VIDEO UNIT, to obtain the checked information B) and C) of the VIDEO UNIT located in the homolog position of the received UNIT but at a distance of a field, using a structure with three memories as indicated in 2c). If the information B) and C) of the received VIDEO UNIT is locally stored in B2', the majority decision is carried out onto the contents of the three memories in the location of the VIDEO UNIT under test equal to the received one. Subsequently the content of the local memory is unloaded into the same location of the memory B3' representing the first repetition, in order to allow the repetition possibility of operations. It is worth noting that the delay necessary to cumulate the three repetitions is here placed in reception, so the schemes in Figures 1a and 3b follow this assumption. However it is possible to move this delay towards the transmission with some little modification of the devices previously described. In this case for example it is advisable to exchange the memories related to the BODY F) between the transmission and receiving sides. The comparison device B6' provides the privileged B) and auxiliary C) information for the VIDEO UNIT under test. The information related to the cell number at variable bit rate (HNC) included in the HEADER E) (VBRH) allows to verify if the quantity of arrived cells is exact and to take a posteriori the majority decision, that therefore is initially applied only to the constant bit rate part (CBRH). The information related to the cell number of the BODY F) (BNC) allows the device B8', that has access to the memory B4' and to the table of pointers B5', to establish if the arrived cell quantity is correct. Moreover the same device can localize the last cells and recognize if the received cells are wrong, using the extracted CRC codes, the BNC and a suitable electric window method. In case of detected errors, it rejects the whole part of the BODY F) covered by the CRC code and generally extended onto a no integer number of cells.</li><li>4c) The unpacking of the VIDEO PACKET is followed by its respective VIDEO UNIT reconstruction, in terms of the prediction errors (S6') and the information associated to the unit. The device B6' provides the privileged information B), such as the selected predictors (S2'), the motion vectors (S3') in case of prediction with motion compensation the quantization step size (S5'), the refresh (S3') and the video format (S1'). The device B7' restores the normal information A) format supplying a signal of valid or not valid data (S7') respectively in conjuction with corrected or lost prediction errors (S6'), here lost errors mean not received or wrong ones.</li></ul>

0025The decoding scheme with error concealment is shown in Figure 3a: <ul id="ul0008" list-style="none"><li>1d) In case of valid data, i.e. exact prediction error the scheme follows the usual structure of a hybrid decoder, conformly to what illustrated at point 4a).</li><li>2d) On the contrary if the data are not valid, the prediction error is not added to the prediciton block (switch B9') therefore the reconstructed block is exactly the prediction block. Moreover the search of the prediction block (in B10') is different from the usual structure given in 1d) for the intrafield selection. As already pointed out, in case of refresh, equal to a forced intrafield the prediction block corresponding to the best selection is supplied, while in case of intrafield coming from the selection, the prediction block is the nearest homologous block. As it does not exist a dialog between the receiving and the transmitting sides this error concealment diversifies the image reconstruction of the transmitter with respect to the receiver. This fact even more imposes the use of the refresh in order to guarantee the identity of working between of the transmission and reception machines; in any case the refresh rate depends on the probability of cell loss that characterizes the network.</li></ul>

Progressive coding and priority networks

0026As a further characteristic of the invention, it is shown that a generalization both of the VIDEO PACKET and of the structure related to the formation of the video information in transmission and to the extraction of the same information in reception, allows advantageously the use of the so called progressive encoders.

0027As it is well-known the progressive encoders allow the transmission of images using following refinements; it must be underlined that the information related to enhancement of the signal must be independent from the information that guarantees a prefixed quality of the reconstructed signal.

0028The progressive coding is implicit in the subband method (see e.g. Karlsson, Vetterli "Subband coding of video for packet networks" Optical Engineering, July 1988); on the other hand in case the cosine bidimensional transform is applied to the image segmented in blocks (for example see Lager, Mitchell Yamazaki "Still picture compression algorithms evaluated for international standardisation" Globecom 1988) the progressive update can be achieved through different mechanisms: <ul id="ul0009" list-style="none"><li>1a) Spectral selection: for each block the low frequency terms provide a row version of the image, which is subsequently refined by sending more and more high frequencies coefficients to get the final image.</li><li>2e) Hierarchical or pyramidal decomposition: the whole image is initially filtered and subsampled, therefore is encoded by ADCT (Adaptive DCT);</li><li>3e) Recursive structure: the difference between the original image and the decoded one coming from the encoder of lower level is recursively coded.</li></ul> In order to describe and simplify the problem, it can be made reference to the conventional structure of a recursive encoder 3e) e.g. with two levels (see Ghambari "Two layer coding of video signals for VBR networks" IEEE Journal on selected areas in communications, June 1989), as indicated in Figure 4.

0029We examine the scheme shown in Figure 4 from left to right side: <ul id="ul0010" list-style="none"><li>1f) the encoder C1 and the decoder D1 of first layer can represent. e.g., respectively the hybrid DCT encoder of Figure 1a with the same meaning for the signals S1 (original block), S14 (reconstructed block), S15 (transmitted prediction error) and the decoder of Figure 3a: the information of normal class A) transmitted and received at this level corresponds to the prediction errors. Other coding and decoding systems belonging to the same group of the one used as example can represent the blocks C1 and D1, moreover they can supply data streams, typically the prediction errors, with constant speed.</li><li>2f) the encoder C2 and the decoder D2 of second level can be implemented, e.g.. either using a simple PCM coding (Pulse Code Modulation) possibly with threshold, that means the transmission only of the values higher than a prefixed threshold, or applying a spatial or temporal DPCM (differencial PCM) or a domain transformation. The outgoing data stream S' 15, generally generated at variable bit rate is equal to a reconstruction error and again it can be classified as an information of normal cell A), but with a lower level than the previous S15. Referring to the structure of the VIDEO PACKET, as precedently described the normal cell A) is re-defined as the class which includes the prediction errors S15 and the reconstruction errors S' 15.</li></ul>

0030Generally speaking and making reference to a generical progressive coding this class includes the normal information A0), A1), ... AN), i.e. the normal class A) is stratified on the level numbers N+1 outgoing from the progressive encoder.

0031Consequently the BODY F) is divided in undersections F0), F1)... FN), each of them having the same characteristics indicated at point F)a')b') of the previous description.

0032Then in the structure of the VIDEO PACKET at the HEADER E) is followed by the BODY F) divided in N+1 BODY F0), F1),... FN), each one including: <ul id="ul0011" list-style="none"><li>a") an alignment word (BiAW = Body i_level Alignment Word)</li><li>b") the information of normal class Ai), equal to the errors related to the encoder of i_th level; e.g. in the case of recursive structure with two levels, the prediction errors for the level 0 and the reconstruction errors for the level 1.</li></ul>

0033In addition the last cell is filled with DUMMY BITS in order to avoid inside the same VIDEO UNIT the propagation of the i_th BODY on the following BODY and the propagation of the VIDEO UNIT onto the contiguous VIDEO PACKETS.

0034Now the HEADER E) includes the information of auxiliary class C) equal to the number of cells making the i_th BODIES (BiNC = Body i level Number Cell), as precedently indicated at point E)e).

0035Obviously if the information of normal class A) of a certain i-th level has a constant bit_rate, the corresponding cell number can be fixed a priori.

0036The protection follows the same procedure applied at point E) e) (BiNC, FPBiNC = First previous BiNC, SPBiNV = Second Previous BiNC) for the information of auxiliary class C). On the contrary the protection on the contents of the BODY F), according to what pointed out in E) f) can be diversified in the different BODIES F0), F1),....FN), both for the presence or the lack of CRC codes, their its type the number and repetition (CRCi = Cyclic Redundancy Check i Level, FPRCRCi = First Previous CRCi, SPCRCi = Second Previous CRCi).

0037The structure related to the organization of the video information in transmission and to the extraction of the same information in reception, proposed when the stratification is not used and shown respectively in Figures 1b and 3b can be easily extended for being adapted to the treatment of information stratified in more levels.

0038In case of recursive structure, the refinement of the i_th level is possible only if the enhancement of previous level (i-th) was successful that is the use of the BODY Fi) involves the correct reception of the BODY Fi-1) belonging to previous level.

0039According to the previously indicated the concealment technique, the error concelment is applied only if some cells belonging to the BODY F0 are lost or deemed wrong using the protection method already pointed out. The following BODIES Fi) are specifically employed to improve the image of the previous level and they can be used only if they correctly arrived and if the information of the previous level was considered valid.

0040As a further feature of the extension it is worth noting that for transmission on packet networks the stratification of the video information implies the generation of cell groups with different priority and therefore the possibility of exploiting advantageously new generations of packet transmission networks e.g. ATM networks, with priority.

0041In the extension of the VIDEO PACKET structure, as was proposed here, the BODIES Fi), with increasing index i, characterize information units and therefore cell groups in which they are packed with a gradually decreasing priority.

0042Obviously the greatest priority belongs to the cells of HEADER E), including the privileged information B) and auxiliary information C) of the present VIDEO UNIT and of two other VIDEO UNITS located in the original image at a suitable space distance.

0043If the priority of the network carries out only a privileged treatment in terms of cell loss probability then all cells of the VIDEO PACKET follow a same virtual path and they are orderly delivered in reception, with a cell loss probability that increases with the i_th order of the BODY Fi), according to the decreasing priority.

0044Figure 5a refers to the case of a single VIDEO PACKET including a single HEADER E) and a BODY F) now divided according to its distinctive feature in N+1 BODIES F0), F1),...FN), equal to the number of stratified levels outgoing from the progressive coding. On the contrary Figure 5b refers to the case in which the priority implies a cell counting on different virtual channels according to the required priority level. Then the VIDEO PACKET P.V, can be divided in a number (M+1) of VIDEO PACKETS P.V.0, P.V.1, .... P.V.M., with M+1 being minor or equal to the number N+1 of levels outgoing form the progressive encoder, so that many BODIES Fi) can be grouped in a single VIDEO PACKET P.V.k, see e.g. P.V 1 including the BODIES F1) and F2).

0045Then each VIDEO UNIT is associated to a VIDEO PACKET subdivided in (M+1) parts: among these packets the VIDEO PACKET 0 includes at least the HEADER E) and the BODY F0), that is all information essential for the basic reconstruction of the video signal. In fact, similarly to the case of a single VIDEO PACKET, the HEADER E) includes the information of privileged class B) and auxiliary class C) common (e.g. VF, FPVF, SPVF) and specifical for each level outgoing from the progressive encoder (e.g. BiNC, FPBiNC, SPBiNC) for each of the three VIDEO UNITS. On the other hand the BODY F0) includes the information of normal class A0), that guarantees a prefixed reproduction quality.

0046It is worth noting that for the used error concealment tecnique the information included in the HEADER E) is the most important one, so it may be reserved a better treatment to the HEADER E) than to the body Fo). Obviously this consideration may be extended to the previous coding without stratification.

0047If the network does not provide an ordered delivery of the information coming from the virtual channels, the receiver structure must further be revised to allow the combination of various information data streams included in different VIDEO PACKETS.

0048Then the generic VIDEO PACKET P.V.k comprising one or more BODIES Fi), must be provided with an identifier of VIDEO UNIT (VUI), in addition to the BiAWs already present, in order to make possible the recognition of the VIDEO UNIT to which the VIDEO PACKET belongs. For example, in case of recursive structure with two levels a VIDEO PACKET subdivided in two stratified PACKETS is associated to each VIDEO UNIT: the VIDEO PACKEt 0 includes the HEADER E) and the BODY F0), i.e. the prediction errors and the VIDEO PACKET 1 that includes the BODY F1), i.e. the reconstruction errors.

0049The invention was described for illustrative aims with particular reference to the fulfilment form represented in the annexed drawings. It must be underlined that this form can be submitted to changes, modifications complements replacements (also partial ones), etc., that being in the reach of the mean skilled technician can be considered as naturally falling within the spirit and scope of the present invention.