WO1991013516A2

WO1991013516A2 - Method for coding video signals

Info

Publication number: WO1991013516A2
Application number: PCT/BE1991/000015
Authority: WO
Inventors: Silvano D'agostino
Original assignee: Alcatel Bell-Sdt S.A.
Priority date: 1990-02-26
Filing date: 1991-02-22
Publication date: 1991-09-05
Also published as: EP0517719A1; WO1991013516A3; AU7235891A; BE1003827A4

Abstract

A universal method, known as VIDEOCODEC UNIVERSEL, for coding video signals for digitally transmitting or memorizing data relating to a still or moving picture, said method being characterized in that: image data is organized by the image memory controller; transformed coefficients are processed by a flexibility processor (FXP); transformed and processed coefficients are reorganized and entropically coded by the Universal variable length coder (U-VLC); the coefficients thereby coded are structured by the universal multiplexer; the coded data are controlled in a buffer by the BRX (Buffer regulator); the regulated data are collected into cells by the packetizer/multiplexer - depacketizer/demultiplexer; the functions described above are reversed by the decoding function; the local clock of the decoder is picked up by the ASD; coding and decoding are made flexible by specifical parameterization of procedures; installing in parallel several VIDEOCODEC UNIVERSELs allows efficient coding of high definition television.

Description

UNIVERSAL METHOD FOR CODING IMAGE SIGNALS

Introduction

The UNIVERSAL VIDEOCODEC described below implements the concepts necessary for coding with reduction of the bit rate of different video services (see patent application EP-88 870 098.6)

The video services considered are of the interactive type and distributed from the high resolution still image, the high quality videotelephone, videoconference, video library, secondary distribution TV, primary distribution TV, contribution TV up to the low resolution part of the TV sub-band of a compatible HDTV transmission.

In addition, by putting several UNIVERSAL VIDEOCODECES in parallel, the contribution and / or distribution HDTV can be effectively coded.

The video interface used in front of the encoder and decoder is of the CCIR 656 type. The video interface of the terminal can be of the PAL, SECAM, NTSC, D2-MAC type or any other composite standard, subject to conversion into composan - your Y; CR; CB defined by CCIR 601.

The network interface used depends on the ATM environment available; STM; G703; SDH; ... Depending on the network used, the channel adapter as well as the network adapter will be determined from the interface of the encoder / decoder proper.

The system described can operate in adjustable fixed rate mode (AFBR) and / or in variable rate mode (VBR) when an ATM network is available. The universality comes from the various video applications envisaged, at different quality levels, with several video interfaces and several types of networks, notably ATM, which allows the operating mode in variable bit rate. 2. DESCRIPTION

2.0 Architecture and functionalities (see Figure 1 encoder; Figure 2 decoder)

The architecture chosen is of the hybrid type: transformed in the spatial domain and differential in the temporal domain.

The CMI (image memory controller) makes it possible to transform the interlaced format of the CCIR 656 into a progressive TV format by merging the 2 frames of the image through the use of a frame memory (FLM).

The DCT transform, outside the loop, operates internally or intra-image.

The FXP (flexibility processor), which implements adaptive linear quantization, makes it possible to carry out the selection of the different quality levels required, by adapting the quantization step. The latter is also adjusted according to the criticality of the coded block as well as the filling state of the output buffer.

The coefficient image memory (FRM) is accessed via a dedicated controller (CMB).

The U-VLC achieves the optimal entropy coding of the quantized coefficients in a universal, that is to say totally audio-adaptive, whatever the type of video sequence. The channel adapter, depending on the type of network used, has a common part which constitutes the coded image buffer accessed through the BRX (Buffer regulation). The latter implements the regulatory laws, the calculation of TF (transmission factors) as well as the channel flow integrals.

The PKX (packetizer / multiplexer) used in ATM implements the AAL layer (ATM adaptation layer) necessary for video as well as sound and data when they are transmitted. in the same virtual channel.

In the case of fixed bit rate networks, plesiochronous multiplexing of the sound and data video components is carried out which allows adaptation to the channel. The optical or electrical line transmission functions as well as the processing of the header in ATM are carried out by the network adapter.

2.1 Image memory controller CMI and frame memory FLM (Field Memory) The CMI provides the interface between a standardized television signal CCIR 656 and the DCT function. It controls the FLM frame memory. He realises:

- decoding of the CCIR 656 protocol

- recovery of auxiliary data up to 128 bytes per frame

- "LINE TO BLOCK" conversion

- nultiplexing of blocks

- the fusion of a complete image from 2 frames

- optional frame deletion - synchronization of other parts of the CODEC

- control of a SRAM memory up to 4 Mbit

- real-time operations (27 MHz * 8 bits).

The data enter the interface R of the CMI which extracts useful video information therefrom, groups them 4 by 4 and component by component to store them via the interface MF in the memory FLM.

Two main modes are to be considered: a) The "DCT INTRA TRAME" mode. In this mode, an algorithm performs the "LINE TO BLOCK" conversion which consists of grouping 8 successive lines of the same frame called "STRIPE". Then cutting this "stripe" into blocks of 8 by 8 pixels and this separately for the 3 components of the television signal. The blocks of the various components are then multiplexed in a programmable order on the interface D. b) "DCT INTRA IMAGE" mode. In this mode, a particular addressing algorithm makes it possible to group 8 lines of which 4 belong to the first frame and 4 belong to the second frame to form the "STRIPE". This is called the merging of 2 frames. Since the frames arrive sequentially, it is necessary to store a complete frame in the FLM memory before starting the grouping.

Given the continuous flow of data, the algorithm in question efficiently manages the memory areas, using an index table to avoid the complete storage of an image. As soon as the merged stripe is formed, as in mode (a), it is cut into blocks of 8 by 8 pixels separately for the 3 components of the television signal. The blocks of the different components are then multiplexed in a programmable order on the interface D. The optional frame deletion consists on the basis of a parameter "S" of sending only the blocks of the first to the interface D frame of a set of successive "S" frames. This functionality is only valid with mode (a). The CMI generates synchronization signals to the other parts of the codec in particular; a signal indicating that it is the first or second frame, a signal indicating the start of a stripe and a signal indicating the start of a block. 2.2 DCT

The bit rate of the image sequences can be considerably reduced when, from a statistical point of view, there is a strong correlation between the elements of visual information. This correlation, also known as redundancy, exists between elements of the same image and between elements of successive images. One category of coding methods includes transform methods.

In these methods, a number of adjacent elements of the image are combined into a partial image called a block. This block is two-dimensional and includes elements taken from a rectangular area of the image, of size MxN.

The next step is to apply to the elements of the block a linear mathematical transformation producing ideally decorrelated parameters and called transformed coefficients.

Among the different existing transformations, the discrete cosine transform (DCT) is one of the ones that has the best decorrelation properties and is one of the most used in image coding.

The transformed coefficients are expressed as a sum of the samples of the image weighted by cosine functions:

M-l N-l

X (, l Σ Σ x (m, n) cos (^^ k π) cos (- ^ - ^ 1 π) ^cl m = 0 n = 0 2M 2N where cl is a normalization factor. 2.3 FXP - CMB - FRM: flexibility processor, loop memory controller, loop memory 2.3.1 Calculation of the quantization step - Adaptive quantization a. The quantization step applied to each transformed coefficient is a function

1) intrinsic properties of the image

2) network capacity b. The link between what can be sent over the network and the calculation of the quantization step is carried out via the "transmission factor", its calculation is detailed in points 2.5 and 2.4. vs. The adaptation of the quantization step to the intrinsic properties of the image is done at coefficient occurrence: no quantification adapted to the order of the coefficient processed (reference 1) at block occurrence: no quantification as a function of the block's "criticality" . d. Criticality concept

Criticality determines the resistance of a block to quantization noise the more critical a block, the more visible the quantization noise.

This notion is not to be confused with one activity that is 1 picture ^• 1 energy contained in the coefficients transformed AC.

A block can be active without being critical, or the reverse, or both, or neither.

The aim is to reduce the quantization noise (by a more appropriate quantification) for the critical blocks to the detriment of the less so blocks, hence the name "adaptive quantization". The consequence of this is a variable quantization noise in the image but a uniform quality in space. In addition, in a codec using a DCT applied to blocks taken within an image, adaptive quantification has a second role to play: avoiding excessive tempo¬ filtering. Indeed, the blocks appearing at the entry of the DCT resulting from the fusion of 2 frames have vertical transitions due to the movement occurred between the taking of the two frames.

"High vertical frequencies" coefficients are then present and must be protected to prevent their elimination from producing a temporal filtering effect. Calculation of criticality

- The criticality calculation is performed on the basis of the transformed coefficients. These 63 coefficients are divided into different zones.

This separation makes it possible to characterize the different structures existing in the Pel * domain of vertical, horizontal, oblique, random structures or with fine details. Each of these structures reacts differently to quantification.

- Each of the zones will be characterized by a parameter: the maximum of the AC coefficients taken in absolute value iijn H ^ΛV -max '

- Certain tests applied to these max ACs and to the differences between these max ACs make it possible to distribute the blocks in 4 classes of criticality. - The criticality of a block must be calculated

• before any quantification

• from the intra block

• on luminance blocks only f. For each criticality class, we will set

- a table of weighting coefficients

- an average level of quantification. g. The choice of a quality level (in the case of VBR coding) directly controls the determination of the transmission factor. 2.3.2 Hybrid coding - Inter / intra modes a. Inter-intra modes

The FXP allows the possibility of performing a predictive coding: the transmitted block is then the difference between the current block and a prediction of this block. This prediction is nothing other than the block positioned in the same place in the previous image.

Such a method requires an image memory which stores the current image to serve as a prediction for the image following the FRM; its control is provided by the CMB. The CMB attached to the FRM has a function identical to a FIFO. b. Inter-intra decision

Each block is left with the option of transmitting the "intra" or "inter" version of the block. The choice will be made judiciously with the aim of transmitting the most economical version in terms of speed.

The chosen criterion estimates, at the output of the DCT, the number of bits necessary for the intra version and the inter version.

This estimate takes into account the fact that the coefficients will be weighted and that each coefficient will not be transmitted with the same number of bits. The estimation of the number of bits is carried out as follows:

2.3.3 PC / DPCM

Predictive processing in the spatial direction is also applied in the sense that the difference between the DC coefficient of the block and the DC coefficient is transmitted from the previous block. This technique makes it possible to take into account the spatial correlation between blocks. 2.3.4 Automatic still picture

A special operating mode ASP (Automatic Still Picture) is activated automatically when the image to be transmitted becomes fixed or weakly animated.

In this mode, the overall quality of the image is gradually improved by relaxing the quantification while maintaining an output rate remaining in the range authorized by the channel.

The final step in this process is the "Freeze" mode where the quantization step is maintained at its lowest value.

2.4 Universal entropy coding 2.4.1 The Universal Variable Length Coder (UVLC)

The transformed and quantized coefficients are encoded by an entropy coder.

In order to be able to obtain a universal entro¬ coding scheme, that is to say independent of the content of the image information, a new coding scheme has been developed.

This scheme, which we will call UVLC, is applied to groups of blocks of transformed coefficients and uses universal codes specific to binary sources without memory.

In one possible embodiment, it is assumed that the size of the blocks is 8 × 8 and that the sampling format is that of CCIR Opinion 601.

A set of 8 lines of the image therefore gives rise to 90 blocks.

Entropy coding is applied to such groups of blocks, rather than to each block taken separately.

The binary representation of the coefficients in this group of blocks is the sign and magnitude representation and will be assumed, to fix ideas, not to exceed 12 bits. This group of blocks can therefore be described as a three-dimensional table of bits. This table consists of: - 64 lines corresponding to the order of the coefficients;

- 12 columns corresponding to the bit levels, from the MSB (Most Significant Bit) to the LSE (Least Significant Bit) plus the sign bit - 12x64 lines, obtained by grouping the 90 bits of the same level with a coefficient of the same order.

The coefficients of a given order are encoded by a run-length coding of their bit lines, from the MSB to their LSB (the sign bit is placed after the LSB): when a bit non-zero is encountered, the other less significant bits of the coefficient to which this non-zero bit corresponds are sent uncoded and the whole of this coefficient is removed from the table.

In this way, the non-zero coefficients are encoded by giving the position of their most significant non-zero bit (MSNZB: Most Significant Non Zero Bit) and by sending the less significant non-coded bits.

The code used to encode MSNZB is a code applying to binary sources without memory any universal code can therefore be used for this purpose (ATRL: adaptive truncated run length code, Lynch-Davisson codes,

Q-Coder, ...).

Generally these codes consist of a prefix and a suffix: - the prefix gives the number of non-zero bits in the line

- the suffix allows to position the MSNZB.

In order to increase the efficiency of the code thus obtained, we can group the coefficients by class and send, for each class, a prefix indicating the highest position for the MSNZB. This avoids scanning bit lines which have a very high probability of being entirely zero.

The implementation of such a scheme is generally done in 2 passes: - first, we locate the positions of the MSNZB, we store the less significant bits of the non-zero coefficients and we count the number of MSNZB per line;

- secondly, we encode the positions of the MSNZB and we sends over the channel from MSB to LSB.

The first pass can be performed on the 12 bit lines of a coefficient in parallel because there is, at most, only one MSNZB per coefficient. The second pass is done bit by bit line from MSB to LSB.

In any case, at most one code word is generated per transformed coefficient.

2.4.2 Universal Video Framing UVF (or Universal multiplexer) L ¹ universal video framing is likely to have a multitude of operating modes in several environments:

1. suitable for several formats and standard qualities defined, high definition, high quality videotelephony, high resolution still picture

2. suitable for different types of networks: STM and ATM

3. suitable for different types of channels: VBR and AFBR (adjustable fixed bit rate.

The universal video framing is divided into two parts:

1 - at frame level

2 - at the stripe level (strip made up of n adjacent lines)

The ^one universal video framing must meet the following func- tions:

1. UVF identification

2. UVF synchronization

3. frame type identification

4. identification of the video system 5. identification of the video format

6. identification of the operating mode

7. system synchronization

8. system configuration

9. UVF protection against transmission errors The details are given below:

1. UVFF Universal Video Framing Field (160 bit)

1.1 FS

FSW Field 48-bit Word synchronization 1.2 FCP Field Coding Parameters IBR Integral Bit rate 18 bits BO Buffer Occupancy 16 bits ST System Type 0 50 Hz 1 bit

1 60 Hz

FS Field Skip 0000 0 (no skipping) 4 bit VF Video Format 0000 4-2-2 4 bit

ST and FS → temporal resolution

VF → spatial resolution DCTF DCT Format 0 8X8 1 bit

1 16X16

.ST. and .VF. and, DCTF. →

Nb: number of blocks per stripe

Ns: number of stripes per field FS Field Status 0 even 1 bit

1 odd

BF Block Format 0 block within a field 1 bit

1 block within a frame

• FS, and .BL. if BL≈l odd field = lst half-frame even field = 2 nd half-frame

PO Parameters Occurrence 3 bits Bit 1: inter / intra 0 block

1 quadblock Bits 2 and: 3 critic. 00 no critic

01 block

10 luminance block

11 quadblock

MC Motion Compensation: 00 without motion compensation 2 bits

01 without me (with hybrid loop)

10 with me (pel accuracy)

11 with me (half-pel accuracy)

MVC Motion Vector Coding 0 non coded 1 bit

1 coded

PI Phase Information (PAL) 12 bit (or frame number) BRMF Bit Rate Mode Flag 0 AFBR 1 bit 1 VBR

VBR Video Eit Rate 0000 34 Mit / sec 4 bit

(16 bit rate levels (AFBR)

16 quality levels (VBR)) WM Working Mode 0 interlaced 1 bit

1 progressive

CN Channels Number 4 bit

→ information if several channels if info splitting ARF Aspect Ratio Format 0 4/3 1 bit

1 16/9

CI Configuration Information provision (37-TBD) bits CRC Cyclic Redundancy Code TBD bits

(applied to the FCP)

2. UVFS Universal Video Framing Stripe (92 to 1307 bits)

2.1 SSW

SSW Stripe Word 48-bit synchronization

2.2 SCP Stripe Coding Parameters

SN Stripe Number 8 bit

BO Buffer Occupancy 16 bit

IBR Full Bit Rate 18 bit

8-bit TFy Transmission Factor (luminance)

8 bit TFc Transmission Factor (chrominances)

SIS Side Info Stripe 0 → 121-5 bit .PO. and .MC. → SIS lenght

MI Mode Identification 00 intrafield block 360 bits

01 interfield

10 interframe

11 intraframe

Criticallity block 360 bit lum. block 180 bit quadblock 90 bit

MV Motion Vectors MVx 6 bits 0 or 495 bits MVy 5 bits or VLC CRC Cyclic Redundancy Code TBD bits (applied to the SCP and VLC)

2.4 3 Error masking at U-VLD

The decoding of the coefficients is done in two time .

While the VLD reconstructs a stripe in the third dimension, it transfers the coefficients of the previous stripe to the FXP ^'1 . This technique introduces a delay of a stripe and allows during error detection:

- either to read consecutively the previous stripe twice

- either cancel the higher order coefficients as soon as an error is detected; detection is then said to be progressive.

2.5 Buffer regulation (BRX)

Buffer control and flow regulation are implemented for both coded variable speed

(VBR) and for fixed rate coded (FBR). They are based on the modeling of digital TV image sequences.

This modeling describes these sequences in terms of entropy information rates as a succession of steps, slopes and flow pulses. These steps are of random levels and durations; they correspond to the occurrences of sequence change (eut off) as well as to the intrinsic complexity of each of the sequences.

On top of this process, a cyclostationary random phenomenon is superimposed corresponding to the internal variability of each image, its period equaling the duration of a frame or that of an image.

The regulation of the codec is adapted to this model: the buffer smoothes intra-frame and / or intra-image fluctuations (short-term smoothing); the regulation is designed to react to sequence changes and their difference in complexity according to time constants programmable via the parameters (long term).

In addition, the responses to the pulses, steps and ramps are calculated to avoid any oscillation of the response (critical or sub-critical damping) so as to avoid visible fluctuations in image quality and achieve locally almost quality -uniform. In FBR mode, the buffer is permanently filled with excess speed from the VLC on the fixed speed sent to the transmission channel; it continually accumulates this excess. A regulation loop makes the buffer state back on the quantization step (FXP) and regulates the occupation of the buffer at a fixed reference level (the setpoint). At the end of the coding of each band of eight image or frame lines (called "stripe"), the loop takes the value of the level of filling of the buffer by the interface M. These levels of filling are compared to the setpoint , the difference is processed by a PID processor which performs both scaling (product), cumulative summation (Integral) and numerical differentiation (Derivative). The sum of these three operations is scaled and provides a quality control factor (TF) which will be applied to the quantizer (FXP) for the calculation of the quantization step during the coding of the next "stripe".

This factor, called the transmission factor, is transmitted to the decoder with the occurrence "stripe". The formulas leading to the calculation of TF are as follows:

I (n) = I (n-l) + BF (n) - BFref (PID summator)

D (n) = EF (n) - BF (n-l) (PID differentiator)

TF (n) = TFI + Cxl (nl) + dx D (nl) + bi (BF (nl) - BFi) k = il with TFi = TFo + Σ b _k (BF _{k + l} - BF _k ) k + 0

n = sampling instant (end of the stripe)

I (n) = value of the summator

BF (n) = buffer occupation level BFref = buffer regulation setpoint

D (n) = value of the digital differentiator

TF (n) = transmission factor c, d, b. = PID parameters

TF ₀ = transmission factor with empty buffer BF _k = limit of the different buffer zones i = index of the buffer zone TF _j = shift of the transmission factor at BFi level

The division of the buffer into different filling zones with bi-different slope coefficients makes it possible to create fallback behaviors to prevent possible overshoots of the maximum capacity of the buffer and the lack of throughput to be supplied to the trans¬ mission.

An automatic filling called "Stuffing" compensates for any deficit flow in the event of an empty buffer. The PID regulator is calculated so as to avoid any excess of the maximum buffer capacity given the most unfavorable initial configurations.

Two transmission factors, one luminance, the other chrominance, can possibly be calculated separately by the PID according to sets of programmable and different parameters.

In VBR mode, the buffer is supplied by the output speed of the VLC in excess of the speed profile negotiated at the initialization of the transmission between the codec and the transmission channel.

The regulation of the filling level of the buffer is normally disconnected except, in the event of a fallback, when the filling level exceeds a programmable threshold in which case, it is restored to ensure a return of the filling level to its setpoint. This regulation works in a similar way to that described for the FBR mode; it is disconnected when the output rate of the VLC decreases and allows a decrease in the level of occupation of the buffer.

The FBR operating mode then appears as a special case of the VBR mode where the flow profile is constant. The limit values of the integrator, the differentiator, the different buffer zones and the corresponding shifts (TFi) of the transmission factor are also programmable. Flow integrals are calculated by the BRX and are transmitted on the occurrence stripe to the resynchronization system of the decoder. Frame or image synchronization is detected on the encoder side (BRX) and serves as a reference for transmission of the full bit rate by the transmission multiplex. At the decoder, the reverse operations are carried out at the BRX input, synchronization is detected and the flow rate integral of the encoder is extracted. A flow integral is calculated analogously at the decoder. These two integrals allow on the one hand to reset the integral of the decoder and on the other hand the calculation of the resynchronization condition

BOE + BOD = l R (t) dt

BOD = filling level of the buffer at the decoder

BOE = fill level of the buffer at the encoder

^* tl R (t) dt = flow integral at encoder 0 t2 (t) dt = flow integral at decoder 0

Error masking at BRX ^"1

The BRX decoder transfers the error from interface B to interface C.

Each data transferred to the BRX is accompanied by a validation signal; this signal is restored to the decoder when said data is read by the UVLD.

The detection of the start of an error controls the memorization of the write address pointer at a FIFO stack (First IN, First OUT); the end of error address is stored by the same principle. When the read pointer reaches the first write address, the error signal is output and the stack is shifted by one.

2.6 Packetizer / Multiplexer and depacketizer / demultiplexer (PKX) The main function of PKX is to assemble cells consisting of a header of 5 bytes and an information field of 48 bytes. This information field can be made up of different types of data (sound, data, video for example). The PKX performs the ATM adaptation layer (AAL) by especially the AAL SAR-PDU and AAL CS-PDU functions.

Video data enters via interface B, is accumulated in a buffer memory in order to be able to fill a cell. On the basis of an authorization signal, the cell is sent to the interface I. The data is accompanied by indications such as: the type of data, their number, the cell number and protection codes against bit errors or cell loss.

The functions of the segmentation and reassembly layer (SAR) are: A. The stubbornness field

The header field occupies the first byte of the ATM cell information field. It contains two sub-fields: 1) SN 4-bit sequence number

The sequence number is used to detect breaks in the cell sequence (loss or insertion). A 4-bit sequence number can detect up to 15 consecutive cell losses. The value 15 can indicate a redundancy cell.

2) The type of 4-bit CT cell

This type of cell is used to indicate:

- a "Continuation of Message" -COM- (CT0)

- an "End of Message" -EOM- (CT1) - a "Begin of Message" -B0M- (CT2)

- a "Single Segment Message" -SSM- (CT3)

- an "RTSW only cell" (CT4)

- an "EMPTY cell" (CT5)

- a "Redundancy cell indicator 1" -RCI1- (CT7) - a "Redundancy cell indicator 2" -RCI2- (CT8)

- active, component (CT11)

- active component + RTSW (CT15)

- CT6; CT9; CT10; CT12; CT13; CT14 remain to be defined. B. The end field The end field occupies the last two bytes of the ATM cell information field. It contains two sub-fields: 1) The 6-bit length indicator The length indicator partially fills the cells (SAR-SDU). It indicates the number of bytes placed from the header. 2) 10-bit correction code (FEC) A protection mechanism against binary errors allows to correct single and double correlated errors. It is based on a BCH code (511.502) in combination with a bet bit. The generator polynomial is (X ⁹ + X ⁴ +1). This cyclic code has a hamming distance equal to 4. The code is suitable for correcting single and double correlated errors. It can also detect certain higher order errors. The position of the BCH code and the parity bit is carefully chosen to avoid ambiguities. The functions of the convergence layer (CS) are:

A) end-to-end video service synchronization

A real-time synchronization word generation (RTSW) mechanism transmits a particular cell to the decoder at regular intervals, independently of the coding structure.

It allows synchronization of the decoder clock with the encoder clock by eliminating the jitter introduced by the various elements of the transmission. A special cell is used to transmit this RTSW. It contains 2 bytes containing an identification code (SID), a sequence number (SSN) and a local de-icing indication (DJT).

At the decoder, using a particular algorithm based on time marks, this synchronization information is regenerated.

B) Cell protection

A cell protection mechanism against cell loss or multiple errors in a cell, uses a parity cell and the aforementioned cell sequence counter. The parity cell calculated on the SAR— SDU field and the cell type and length indicator subfields is generated as soon as n * 16-l cells (n positive integer) are sent to the interface I. At the decoder on the basis of the sequence counter, the parity cell and the bit correction code (FEC), detection and / or correction of binary erroneous information is carried out, detection and / or correction and / or recovery lost cells. 2.7 NAC: ATM network access control

This function is used when the VIDEOCODEC UNIVERSEL is connected to an ATM network.

Its purpose is the conformation and limitation of the speed profile used with the parameters negotiated during the establishment of the link.

In practice, this is a preventive police function corresponding to the police function exercised by the network at the access point. Different algorithms can be considered to perform this function: Leaky buket, Gaussian gabarit, peak cell rate, peak rate mean rate, ...

By considering the characteristics of the source as well as the constraints of the network, the ideal algorithm can be chosen and its parameters determined.

The characterization of the source, as well as the constraints of the network, are not well known to date; therefore, this function remains to be specified.

However, its mode of action is determined and consists of limiting requests for access to the "I" interface, by not authorizing the PKX (packetizer) to access the BUF (Buffer), which in turn by the the BRX (Buffer regulation) mechanism will react if necessary through the FXP (flexibility processor) on the quantification of the source in order to control the flow.

2.8 TCM ATM communication module

Its purpose is the implementation of the physical and ATM layers, respectively the line termination and header processing functions (HEADER) of ATM cells.

Network / terminal signaling is achieved through to this module, located in the network adapter.

2.9 ASP: disengaging the synchronization in ATM

This function is used when the VIDEOCODEC UNIVERSEL is connected to an ATM network, in line with the decoder.

Its purpose is the implementation of the adaptive PLL (phase-locked loop) algorithm, the function of which is to process the RTSW (real-time synchronization words) coming from the depacketizer / demultiplexer) (PKX ^"1 ) in order to reconstruct a local clock at 27 MHz attached to the source clock also at 27 MHz, this ensuring stability of the local clock in accordance with CCIR ^' 656 recommendations notwithstanding the cell jitter introduced by the ATM network. Adaptive must also hang on and achieve its performance as quickly as possible.

2.10 System aspects and codec implementation

1) The functions CMI, DCT, FXP, CMB, U-VLC / U-VLD, BRX and PKX are realized in the form of electronic circuits with very high integration density.

2) Reversibility of functions

Each of the functions mentioned above is reversible. That is to say, they can perform the corresponding function in encoder mode or in decoder mode. They also carry out operations which, depending on the mode, are either useful for the encoder or useful for the decoder, in which case the operation in question is deactivated in the reverse mode.

3) Distributed time base Each of the aforementioned functions except the PCT has devices for sequencing data over time, thus enabling each function to provide its operators with the appropriate time signals derived from the common clock at 27 MHz. These devices are programmable. Pes signals are transmitted between these different sequencing devices in order to ensure the synchronization of a function with respect to the previous one. 4) Bus and flexible configuration and control procedures for coded (X3C)

Each function has an X3C serial interface for configuration and control of the codec. This interface has two bidirectional lines: a line for the clock signal and a line for the data signal.

The exchange of data between two modules is based on a protocol allowing the transmission of information relating to the codec. This information is either elementary binary information (1 bit) or a group of n binary values (register) or a set of registers (tables). The protocol consists of commands accompanied by addresses and / or data of different lengths. The protocol is oriented towards a dialogue between: a configuration processor and each function of the codec. A transmission between two functions controlled by the processor makes it possible to speed up exchanges. The commands implementing a single bit are composed of a single byte containing the address and the data. Several orders of this type can follow one another. The commands implementing a register are made up of 2 bytes: the first contains the address, the second the data. The following exchanges allow access to neighboring registers (automatic incrementation of the address).

The commands implementing a table are composed of at least 3 bytes, the first contains the number of the table, the second the access position relative to the start of the table, the third the first data relating to the position d 'access. The following exchanges allow access to the neighboring registers of the table (automatic incrementation of the address). This flexible protocol makes it possible to configure the coding parameters making it universal. 5) Reversibility of the codec.

The functions and architecture are chosen to ensure codec operation either in encoder mode or in decoder mode. Pe more, it requires only one coding (or decoding) chain for the 3 components of the television signal Y, CR, CB operating at 27 MHz. Masguage strategies

Since the transmission of coded information can introduce errors, devices can intervene at different levels to suppress or mask the effects due to these errors. The AAL layer among others has devices capable of correcting transmission errors within certain limits. When their possibilities are exceeded, the data sent to the following functions may be subject to errors. The foregoing devices may or may not detect these errors. If the detection has not taken place, it can still be done using the protection and / or detection mechanisms of the U-MUX. In all cases where detection has taken place, at a given location, with no possibility of correction, masking actions can be taken, these are the following techniques:

- masking by displaying the previous stripe of the same image in the U-VLC or the CMI - progressive masking by canceling the coefficients of higher order than that of the error in the U-VLC

- masking by displaying the stripe of the previous image in the FXP

- masking by displaying a fixed programmable level in the FXP

- masking by displaying the previous stripe of the same frame in the U-VLC or the CMI

Claims

1. Universal method of coding image signals for the digital transmission or storage of information relating to a fixed or mobile image, known as UNIVERSAL VIDEOCOPEC, characterized in that:

- image information is organized by the CMI

- the transformed coefficients are processed by a flexible processor (FXP)

- the transformed and processed coefficients are reorganized and coded entropically by the Universal-VLC

- the coefficients entropically coded by the U-VLC are structured by the multiplex U (U-MUX)

- the encoded data are managed in a buffer by the BRX

- the regulated data are assembled in cells by the PKX - the functions described above are reversed for the decoding function the local clock at the decoder is retrieved by the ASD

- the encoding and the decoding are made flexible by a specific parameterization of the procedures - the paralleling of several UNIVERSAL VIDEOCODECs makes it possible to effectively code HDTV.

2. Method according to claim 1 characterized in that said UVLC encodes the transformed and quantized coefficients by grouping several blocks in a table, and by giving the position of their MSNZB in this table while leaving the less significant bits uncoded.

3. Method according to claim 2 characterized in that the positions of the MSNZB are coded by any universal code provided for binary sources without memory. 4. Method according to. claim 2 characterized in that the encoding is carried out in two passes:

- during the first, the MSNZBs are located and denominated and the less significant bits are memorized

- during the second, the MSNZB positions are encoded and sent on the channel, as well as the corresponding less significant bits, proceeding from MSB to LSB.

5. Method according to claim 2 characterized in that the coefficients of a table are grouped by classes and that a prefix is sent indicating the highest position of the MSNZB of these classes.

6. Method according to claim 2 characterized, also, in that it is applied to sequences obtained by sub-band filtering and no longer to transformed coefficients.

7. Method according to claim 1 characterized in that said U-VLC ensures error masking according to two techniques: - display of the previous stripe of the same image

- progressive error detection and cancellation of higher order coeffi¬ cients.

8. Method according to any one of claims 1 and 2 to 7 characterized in that said U-VLC can be applied to any number of blocks grouped in one or more spatial and / or temporal directions.

9. Method according to claim 1 characterized in that said multiplex U

1. is suitable for several formats and qualities: standard definition, high definition, high quality videotelephone, high resolution still picture;

2. is suitable for different types of networks: STM and ATM;

3. is suitable for different types of VBR and AFBR channels (ajus¬ table; fixed bit rate). The multiplex U is divided into two parts:

- the frame level (UVFF)

- the stripe level (UVFS) breaks down each of its levels into two parts:

- sync word (SW). - coding parameter (CP) _ i. Method according to claim 1 characterized in that said multiplex U comprises synchronization words independent of the words generated by the coder (VLC); this is guaranteed by the addition of a data mixer applied to:

- frame coding parameters (FCP)

- stripe coding parameters (SCP)

- the words generated by the VLC; data mixing and protection of these data against errors being carried out in a single operation.

11. Method according to any one of claims 1 and 9 characterized in that said U-VLC ensures the resynchronization strategy of the decoder according to the verification of the equation BOE + BOD = IBRD - IBRE

BOE = filling level of the buffer at the encoder BOD = filling level of the buffer at the decoder IBRE = integral of the output bit rate at the encoder IBRD = integral of the input bit rate at the decoder in both VBR and FBR; this equality relation tolerates a certain margin of error, programmable, introduced by all of the elements located between the encoder BRX and the decoder BRX. 12. Method according to any one of claims 1 and 11 characterized in that said U-VLC ensures the resynchronization strategy between coding and decoding since several universal codecs are used as elements for coding and decoding part of a high definition television signal or a video signal format other than the standard CCIR 656 or CCIR 624 formats.

13. Method according to claim 1 characterized in that said CMI comprises a particular addressing algorithm using an index table which allows the merging of two frames of the television signal by effectively managing the memory areas and thus avoids storage of a complete picture.

14. Process. according to claim 1 characterized in that said CMI comprises a device which makes it possible to choose by programming the output order of the blocks of 8 * 8 pixels of the different components of the television signal Y, CB, CR.

15. Method according to claim 1 characterized in that said CMI in decoder mode comprises a device which ensures the masking of error originating from the preceding functions of the decoder, by display of the preceding stripe of the same image or of the same frame. 16. Method according to claim 1 characterized in that said FXP performs the criticality calculation in the following way:

- distribution of the 63 AC coefficients in several zones - calculation of the maximum of the coefficients of each zone taken in absolute value ap ^~ p ^~ elé "AC_m ,, a„ x "

- some tests ap ^~ p ^~ liq ~ -ues on these ACmax and on linear combinations of these AC divide the blocks into 4 classes of criticality. 17. Method according to claim 1 characterized in that said FXP performs the calculation of the quantification step as a function of two parameters: nq and nW, each function of different parameters:

- nq = function (criticality class, transmission factor, component)

- nw = weighting factors (see reference I)

= function (order of the coefficient, criticality class, transmission factor, component) nw being chosen from a certain number of tables. 18. The method of claim 1 characterized in that said FXP performs the inter mode / intra mode decision by comparing the estimate of the cost of the block in intra mode and the cost of the block in inter mode; the estimation is carried out as follows: Estimation =

8 8

Σ Σ ^{2 +} M ^log 2 (rshift (blockt (k, l), 2 * _- _f)) - λ-1 k = 1 1 = 1 ^JLJ

where blockt (k, l) is the block of coefficients transformed at the output of the DCT.

19. The method of claim 1 characterized in that said FXP has the option "ASP" (Automatic Still Picture * '); the FXP has three operating modes:

- normal mode

- still mode - freeze mode; switching from one mode to another is automatic or manual; the ASP assesses whether the FXP can switch from normal mode to still mode by counting the percentage of blocks checked in • inter mode; when the still mode is activated, the quantification step is progressively refined, the ASP constantly checks whether the return to normal mode is not required; when the quantization step has reached its minimum value, the Freeze mode is then activated; in this case, all the coefficients pass into inter and are zero; it is therefore decided to inhibit memorization in the loop memory and to allow a certain percentage of blocks to pass intra; this device allows the incrustation of part of the image; the ASP is constantly on the lookout to decide if switching to another mode is not required.

20. The method of claim 1 characterized in that said FXP in decoder mode includes a device which provides error masking from previous functions of the decoder, by displaying the stripe of the previous image or by a programmable level of erro data ¬ born. 21. Method according to claim 1 characterized in that said BRX calculates the values of the transmission coefficient TF by one or more PID control blocks with programmable parameters making it possible to adjust the consta¬ ntes of response time, to avoid overshooting the capacity of the buffer, to ensure an almost uniform image quality over time and to work in VBR or FBR mode.

22. Method according to any one of the claims 1 and 21 characterized in that the BRX provides a regulation based on and adapted to an image modeling described by cycloεtationary random processes superimposed on density variations of very characteristic entropy information: pulses, steps and ramps.

23. Method according to any one of Claims 1 and 21 characterized in that the BRX of the encoder in VBR mode accumulates in the buffer the excess bit rate of the VLC compared to the bit rate profile negotiated at the initialization of the transmission between the coded and the transmission channel and that buffer regulation is initiated in a fallback situation when the buffer occupancy level exceeds a programmable threshold.

24. Method according to any one of claims 1 and 9, characterized in that the BRXs of the encoder and decoder calculate flow integrals in FBR and VBR mode; the flow integrals calculated at the encoder allow attempting to reset the integrals calculated at the decoder; the encoder and decoder integrals contribute to the process of resynchronization of the encoder on the decoder. 25. The method of claim 1 characterized in that said BRX ensures the transfer of an error from interface B to interface C through the use of an error stack; this stack memorizes the write addresses of start and end of error and authorizes the accumulation of several fields of erroneous data.

26. The method of claim 1 characterized in that said Pk implements in the AAL-CS layer an RTSW independent of the structure of the assembled message; the RTSW is structured so as to compensate for the multiplexing jitter (DJT) and makes it possible to protect against losses of synchronization cells (SSN); in addition, its occurrence is such that it guarantees clock recovery at the decoder conforming to the specification of CCIR 656 with maximum efficiency (overhead, duration); in addition, thanks to a mechanism for indicating the arrival of cells at the decoder by time, the jitter introduced by the buffer of n cells necessary for the decoder for recovering lost cells (see claim 28), is fully compensated; this allows the device to function perfectly in the VBR mode.

27. The method of claim 1 characterized in that said PKX implements in the AAL-SAR layer, an FEC based on a BCH code (511, 502) and a parity bit, which makes it possible to correct up to a correlated double bit error, as well as the detection of multiple bit errors. 28. Method according to any one of claims 1 and 27 characterized in that said PKX implements, in the AAL-CS layer, a protection mechanism against cell loss and multiple bit errors in burst or isolated, the mechanism consisting of a vertical parity cell calculated on transmission on a cell block (n) and slid into the flow of cells of a cyclic cell counter (SN); of an FEC (see claim 27), as well as on the calculation and operation of parity cells local to the decoder, which make it possible, by comparison with each other and / or with the parity cell transmitted by the encoder, to correct single and / or burst multiple bit errors.

29. Method according to any one of claims 1 and 26 characterized in that said ASD associated with the operation of the RTSWs compensated by the PKX ^"1 (DJT and time indication), allows by the implementation of 'A PLL adapta¬ tive rapid convergence towards the clock specifications of the CCIR 656, this in the presence or not of cell jitter introduced by the ATM network 30. Method according to claim 1 characterized in that the said specific para-erisation procedures are performed by the use of a flexible X3C configuration and control protocol allowing the transmission of elementary binary information (1 bit) or binary information group (register) or register group (table).

31. Method according to any one of claims 1 and 30 characterized in that said specific parameterization of the procedures implements programmable and distributed time bases capable of generating all the time signals necessary for each function and for several systems or part of television time systems. 32. A method according to any one of claims 1 and 12 characterized in that the paralleling of several UNIVERSAL VIDEOCODECs makes it possible to effectively code an HDTV image cut horizontally and / or vertically and / or temporally.