EP1829377A1 - Method for a variable bit rate transmission through a transmission channel - Google Patents

Abstract

The invention relates to a method for transmitting an audio and/or video program through an adjustable variable bit rate transmission channel consisting in adjusting at least one encoding and/or transmission parameter according to at least one instruction vector to at least one dimension representing a reception quality desired by an end user.

Description

METHOD OF TRANSMITTING VARIABLE BIT RATE THROUGH THROUGH

A CHANNEL OF TRANSMISSION.

Operators involved in the distribution of services including video must provide the end user a level of quality given at the terminal so as not to devalue the content of the service.

They should also minimize storage and / or service delivery costs by controlling a used rate-reducing video coding method, as well as allocated transmission network resources to carry the video service. These coding and transmission methods have a variable impact on the quality returned to the user according to their configuration and the content of the video service.

In addition, the development of digital technologies has resulted in the provision to the public of a wide variety of terminals capable of rendering video images. These terminals have very diverse capabilities, ranging from the small screen of a portable terminal to the big screen of a TV.

There are many known resource allocation methods for rate reduction coding or digital transmission. The differences between these methods lie in the quality measures used, the means of action on quality (ie the resources on which the methods operate), sometimes the optimization algorithms implemented.

There are mainly three known types of quality measures:

Measurements at the physical level result in information relating to basic information units such as bit error rate (BER) or bit error rate (BLER) in the transmission channel, the channel rate, the bandwidth, the transmitted power, the Signal / Interference (SIR) ratio, etc. The bit error rate (subsequently referred to as BER) is the most widely used measure in the literature for characterizing channel errors.

Measurements at the network / transport level are generally related to more structured information elements such as packets: packet loss rate, payload, transmission delay, transmission delay variation. - At the level of the decoded image, objective quality measurements are sometimes made, but they are not used to achieve resource allocation. Video image complexity measurements are also used in statistical multiplexing methods (see L. Bôrôczky, "Statistical Multiplexing Using MPEG-2 Encoders", IBM J.Res., Vol 43 No. 4 JuM 999). , in order to adjust the flow.

These measures are exploited to control various mechanisms of action on the coded video stream:

At the rate reduction coding level: the coding rate, the number of layers in the case of a scalable encoder.

At the interface between the encoder and the network, and in the network: the retransmission of an incorrectly received or received packet, the modification of the level of data protection against errors by correction mechanisms, the differentiated protection data according to their importance, the priority of data transmission, the power of emission. There are two types of resource allocation or optimization processes:

Binary processes engaging an action or allocation of a resource from an event signaled by one of the measures. An example is the triggering of the retransmission of a packet by the network when it is reported missing or erroneous by the receiver.

The methods of piloting a resource from a law determined in a logical or empirical manner. An example of this approach is illustrated in 3GPP Technical Specification 23.107

V5.12.0, "Quality of Service (QoS) concept and architecture", 3rd

Generation Partnership Project: The transmission parameters of a service on a UMTS network are defined according to the type of service.

One particular area that uses resource adjustment techniques is digital television. Today, the adjustment of the coding rate does not take into account the characteristics of several types of possible terminals, and is at least partially manual:

Operators can decide the coding rate for each program according to its content and possibly adjust it manually based on feedback. Alternatively, the use of variable rate reduction coding and statistical multiplexing methods makes it possible to obtain a variable bit rate depending solely on the content of the video. However, it is still necessary to manually set a minimum flow rate and a maximum allowable flow, chosen according to the content of the program.

Operators involved in the distribution of video services should provide the end user with a given level of quality, while minimizing the costs of storage and / or routing of the service, which involves adjusting the use of resources . As shown above, there are at least partly manual techniques and automatic resource allocation techniques. All these known techniques have at least one drawback.

Many coding resource adjustment techniques are now at least partly manual. For example, digital television operators decide the coding rate for each program or program type and implement this setting manually. In practice, very animated programs such as sports broadcasts require a higher speed than others. Alternatively, the use of variable rate reduction coding and statistical multiplexing methods makes it possible to obtain a variable bit rate depending on the content of the video. However, it is still necessary to manually set a minimum flow rate and maximum allowable flow. Moreover, the quality criterion used to adjust the coding rate is a parameter of complexity of the image, and not a measure of quality perceived after coding. Finally, this method does not take into account the characteristics of the terminal to adjust the coding parameters.

The existing techniques for automatic allocation of transmission resources are all based on transmission quality measurements at the network level. However, this type of measurement is not well representative of the perceptual quality returned to the user. It follows from the use of these non-perceptual measures an unsecured quality restituted to the end user, and therefore a non-optimal use of transmission resources. This prevents an operator from guaranteeing a given level of perceived quality, and from using the transmission resources optimally. The present invention provides a method and system for selecting the rate reduction video coding configuration and resource allocation at the transmission network.

The goal is to restore a given video quality level at the terminal and optimize the use of storage resources and / or transmission. For this, the method combines techniques for measuring video perceptual quality and, where appropriate, vector quantization optimization.

The present invention provides a method and system for selecting the rate-reducing video coding configuration as well as the allocation of resources at the transmission network level based on the perceived quality at the terminal and optionally the characteristics of the transmission. terminal of the user.

The goal is to restore a given video quality level at the terminal and optimize the use of storage resources and / or transmission.

For this, the method combines techniques for measuring video perceptual quality and optimization. Measurements of perceived quality can be obtained from the decoded video images, and not from the compressed video stream.

The invention thus relates to a method for transmitting a variable bit rate video program through a transmission channel, characterized in that it implements an adjustment of at least one coding and / or transmission parameter. function of at least one setpoint vector with at least one dimension representing a reception quality desired by said end user.

A said transmission parameter may be the bit rate and / or the type of modulation and / or the transmission power.

Said adjustment is made from a deterministic relationship between the desired reception quality and the encoding and / or transmission parameter (s).

Alternatively, said adjustment is implemented as a function of a distance between said target vector and a measurement vector representing said quality of reception measured at said final user. Said reception quality may be measured on a fixed duration sequence of said program. In particular, said adjustment is made by modifying the transmission power P as a function of a distance between the reference vector and the measurement vector. In any case, said adjustment can be made also according to at least one parameter of the content of the program. A parameter of the content may be an activity parameter and / or a parameter assigned to the name of the program and / or the type of program, said adjustment may also be made according to a characteristic parameter of the terminal.

A characteristic parameter of the terminal may be the resolution of an image displayed on said terminal and / or the bandwidth.

The method can implement the development of a dictionary from a learning set comprising NZ vectors R characterizing the NZ test data, each Rz vector (Z ranging from 1 to NZ) of a rank test z, resulting from the union of a vector Qz representing the perceived quality of this rank test z, and a vector Pz representing the coding and / or transmission parameter or parameters of this rank z test, and possibly a vector Tz representing the parameter or parameters of the terminal of this rank test z, and / or a vector C _z representing the content parameter or parameters of said program.

According to a first variant applicable to the case where the number NZ is not very high, the dictionary is constituted by the vectors of the training set. It consists of a group of N vectors (N = NZ). The maximum NZ number of vectors for which this variant is applicable is strongly dependent on the characteristics of the application (for example the number of search queries for the optimal vector per second) and implementation constraints (for example the computing and of memory that it is possible to assign to the search process of the optimal vector). For example, if it is desired to make only 10,000 vector comparisons per second, it will be possible to perform only 100 optimal vector searches per second in a list of NZ = 100 vectors, or only 50 optimal vector searches per second. second of NZ = 200 vectors. Otherwise, the dictionary is obtained by a vector classification algorithm from said learning set, and is consisting of a group of N vectors (with N <NZ) having a minimum mean distortion with respect to the NZ vectors of the training set. The number of N vectors of the dictionary to be used is strongly dependent on the characteristics of the application and the implementation constraints (as well as for the choice of NZ for the first variant), but also on a compromise between the precision of the dictionary and its size. The larger the dictionary, the more accurate it is, which gives the system superior performance. In practice, a dictionary of N = 20 to 40 vectors is suitable for a training set of 10 different coded sequences with two different resolutions and 10 different rates (ie NZ = 200 configurations).

After constitution of the dictionary, the adjustment can be made by determining by vector quantization the dictionary vector corresponding best to a constraint vector representing at least the desired quality.

The constraint vector may consist of the union of one of the vectors representing the desired quality with a vector representing at least one content parameter and / or a vector representing at least one parameter of the terminal. According to another variant not implementing vector quantization, but involving a measurement at the terminal of the quality perceived for the transmitted program, the method is characterized in that a said adjustment for example of the transmission power P , is performed by step DP according to the difference between the measured perceived quality Q and the target quality QC for the video program.

The process can then be characterized in that:

- If | Q - Qc | is less than a first threshold, the power P is not modified

- If | Q - Qc | is between the first threshold and a second threshold, the power is increased or decreased by the step DP according to whether the sign of Q - QC is respectively negative or positive

- If | Q - Qc | is between the second threshold and a third threshold greater than the second threshold, the power is increased or decreased by kDP with 1 <k = 2 depending on whether the sign of Q - Qc is respectively negative or positive. Advantageously, it is characterized in that the step DP is variable according to a type of content associated with the video program.

Other features and advantages of the invention will appear on reading the description below, in conjunction with the drawings in which:

- Figure 1 illustrates the general context of the provision of video services;

- Figure 2 illustrates a relationship between the coding rate and the perceived quality, depending on the content; FIG. 3 illustrates a gain in flow rate with reference to FIG. 2;

FIG. 4 is an illustration of the method according to the invention;

FIG. 5 illustrates a vector quantization coding, while FIG. 6 illustrates the procedure for constructing a dictionary; FIG. 7 illustrates the dictionary development method in the case of the present invention;

FIGS. 8 and 9 illustrate the search sub-steps of the coding and transmission configuration in the case of a vector quantization; FIG. 10 illustrates the search procedure of the coding and transmission configuration in the case of the use of a deterministic law;

- Figures 11 and 12 illustrate the constitution of a dictionary respectively without classification and classification; FIG. 13 illustrates the selection of the optimal coding rate as a function of the resolution of the terminal and the quality requested;

- Figure 14 illustrates the sequence of adjustment steps to an optimal configuration;

- Figure 15 illustrates the adjustment of the transmission power according to the measured quality and the type of content.

FIG. 16 illustrates, for example, the impact of packet losses on the proportion of lost video images, depending on the type of sequence (slow or fast).

The quality of a video service rendered at the end-user level is significantly influenced by the coding process at rate reduction, the resources allocated to this service in the transmission network, and the capabilities of the display terminal.

Figure 1 shows the main elements involved in the provision of a video service, namely video compression (or rate reduction coding, transmission to the terminal by the transmission network, and finally the terminal).

1) Video compression or rate reduction coding methods:

They make it possible to adapt a stream of binary information representing the video images to the capabilities of the equipment located downstream: network, terminal. But these methods introduce losses of information: the images restored after decoding are not identical to the original images. This can result in visible degradations on the decoded images, which has an impact on the quality of the service delivered to the end user.

The importance of the visibility of coding impairments varies according to many parameters: the content of the video signal, the bit rate of the coded bitstream, the spatial resolution, the refresh rate of the images, and so on. In order to restore a desired level of quality, the parameters of the rate reduction coding method must therefore be carefully selected.

2) The transmission of the bit stream resulting from the rate reduction coding to the terminal by a transmission network:

This transport may be accompanied by loss of binary information. The methods of reception and decoding of the stream at the terminal then restore video signals that can be affected by visible impairments, which has an impact on the quality of the service rendered at the end-user level.

The importance of the visibility of transmission impairments varies according to many parameters: video signal content, bit rate or transmission power allocated, transmission protocol (per packet, with or without a correction method, etc.), distribution and importance of losses, type of information lost, etc. The invention proposes to maintain the level of quality returned to the user, while minimizing the use of network resources by adjusting the parameters of the transmission to the quality required or the quality measured in comparison with the requested quality. 3) The terminal:

The characteristics of the bitstream and the video must be adapted to the processing and display capabilities of the display terminal. For example, it is unnecessary to send a video stream of resolution greater than the resolution of the screen of the terminal, or which requires calculation capacities exceeding those necessary to receive or decode the stream. The characteristics of the terminal therefore constitute constraints to be taken into account when choosing the parameters of the video compression method. The selection of the parameters of the rate reduction coding method relative to a quality level according to the invention enables the video service provider to commit to perceived quality. In addition, this selection taking into account the characteristics of the terminal, it allows the operator to minimize the resources required for storage and / or transmission of this service.

The adjustment of the transmission parameters makes it possible to adapt to a change in the characteristics of the transmission channel in order to maintain the quality perceived.

The invention makes it possible to obtain significant flow rate gains. Indeed, the quality perceived at the end of a rate reduction coding is highly dependent on the coding rate. The type of content and in particular the presence of movements and fine details in the scene require a greater flow rate than a non-animated scene (so-called less complex) to obtain a given level of quality. Figure 2 shows the perceived quality variation for three sequences I, II, III of increasing complexity as a function of coding rate.

Without the proposed resource allocation method and implementing a perceived quality metric, there is no way of knowing the quality being rendered from measurements made at the network level, such as video flow rate measurements. , or bit error rate. A known solution to obtain a good quality is then to allocate whatever the sequence the flow necessary to guarantee the quality of the most complex sequence. In this case, FIG. 3 shows the gains in flow compared with the "III" sequence of FIG. 2, which shows a very significant gain in flow between 20% and 50%. The sensitivity of a video stream transmitted over a digital network varies according to the type of video content. The presence of motion greatly influences the visibility of the damage caused by transmission errors. In the case of transmission over an IP (Internet Protocol) network, it can be observed that for the same number of lost IP packets, the drop in quality is greater for video sequences having a high-motion content.

This can be exploited in practice in a method of adjusting the transmission power of a UMTS transmitter to give priority to "complex" video streams, that is to say with high movement.

FIG. 4 illustrates an exemplary system according to the invention. It is essentially:

A device 1 for measuring the perceptual quality of a video signal in a transmission or broadcasting network. This equipment makes its measurements from the video signals decoded by the terminal. It can possibly be integrated into the terminal.

An equipment 2 for optimizing the coding and / or transmission parameters P, based on the knowledge of the type of video content C, the characteristics of the terminal, the perceived quality QC to obtain (or target quality), and the perceived quality Q actually measured.

- The parameter optimization equipment consists of a DB database entity and a decision entity RECH.

The applicable video perceptual quality measurement methods are those that exploit data from the video decoding process:

or only the pixels of the video images received after transmission (said method without reference),

or the pixels of the video images received after transmission and a small proportion of information of the source images (said method with reduced reference).

Reference will be made in particular to the Patent Applications filed by TELEDIFFUSION DE FRANCE and published under the numbers EP 1020085 and PCT WO 2004/047451, the latter being entitled "Method and system for measuring the degradations of a video image introduced by a coding to flow reduction "for an example of these two types of processes. Full reference quality measurement methods are not applicable because they require the pixels of the video images received after transmission and the pixels of the images before transmission.

The purpose of the optimization procedure is to control the use of resources by searching for a coding or transmission configuration to achieve a given level of perceived quality. One or other of the following two techniques can be used:

1. The exploitation of a database of cases representative of the relationship between the perceived quality and the coding configuration or the transmission network. A vector quantization search engine of the database case that best matches the desired perceived quality under current (imposed) conditions and minimizes the resources requested from the network.

2. Calculation by a logical or empirical law determined in advance, giving the relation between the perceived quality and the coding configuration or transmission network considered.

The optimization procedure can be performed by vector quantization.

Vector quantization is a technique that associates a point X (or vector) of a space with t dimensions at the nearest point

U _k = QV (X), in the sense of a distance Δ, from among a set of N vectors

U _{1 N} called dictionary.

U _h . _N = (Uj J = 1 ... N). Eq. 1

QV (X) = IIA (X, U _i ) ≤ A (X, U _k ); k = l..N Eq. 2A (X, U) being the distance between the vectors (X, U) Eq. 3

This technique for modeling complex processes has for example been used in image coding. The image is previously subdivided into subsets such as rectangular blocks of pixels, the vector quantization is to search for each block of pixels the block of pixels of the dictionary (called vector) closest. Only an index or address of the vector is transmitted to the decoder of the image, decoder which reconstructs the image thanks to the knowledge of the dictionary and the identifiers of the corresponding vectors.

Figure 5 shows the principle of the principle of encoding and decoding by vector quantization. X is the vector to be encoded, U _k are the elements of the dictionary; with k = l..N, N number of vectors. Coding by vector quantization makes X the index (i) of its nearest neighbor in the dictionary. This index is the code word that will be transmitted.

The notion of distance or distortion between two vectors is introduced for the nearest neighbor search in the dictionary. Several distances have been proposed to optimize the vector quantization and to approach as much as possible the fidelity of the initial signals.

The distance or distortion called quadratic error, is among those most used for vector quantization.

(A, B) two vectors of dimension t.

The use of the vector quantization technique involves two main interrelated steps: 1. The formation of the dictionary from a set of learning

2. search for the nearest neighbor using an appropriate distance.

The manner in which these two steps are used in the invention to control the perceived quality of a video service coded by rate reduction and transmitted numerically are successively described in the remainder of this document.

The development of the DB dictionary constitutes a step prior to any optimization of the coding and transmission configuration by vector quantization. The dictionary is a database DB which contains representative cases U _k = UL.N of the relationship between the perceived quality and the coding configuration or the transmission network for certain characteristics of the video content and the given terminal.

In order to develop this dictionary, a set of tests must be performed. The data characterizing these tests constitute a learning set {R _k }, which is used by a specific procedure for constructing the dictionary (FIG. 6). This method is an empirical approach to modeling by learning the relationship between quality perceived and the encoding configuration or the transmission network for certain characteristics of the video content and terminal set.

Each of the NZ tests is identified by its z number. Each test gives a particular case of the relationship between the measured perceived quality Q ₂ and the coding and transmission parameters P ₂ for the characteristics of the terminal T ₂ and the video content C ₂ given. The choice of the different tests carried out leads to a powerful dictionary. For this, in order to allow a good modeling of the relationship between these different parameters, the parameters P ₂ , T ₂ and C ₂ are varied on the one hand over a range corresponding to the operating conditions in practice, and on the other hand in order to obtain the desired perceived quality levels Q ₂ (FIG. 7). Qz, Pz, T ₂ and C ₂ are vectors in the most general case: Q ₂ = (VQ _l2 , .., VQ _{nq> 2} ) Eq. 5 with nq: number of quality parameters and VQi .. _nqι2 : quality parameters for the z-test

P _Σ = (vP _lz , .., VP _ιφ J _Eq.6 with np: number of coding and transmission parameters and VPi .. _nPιZ : coding and transmission parameters for the z-test

T ₂ = (VT ₁₁ , .., VT _n J Eq 7 with nt: number of terminal parameters and VTi .. _nt , ₂ : terminal parameters for z test

C ₂ = (VC ₁₂ , .., VC _n J Eq.8 with ne: number of parameters of the content and VCi .. _nc , z: content parameters for the test z Each learning vector R ₂ of dimension t is resulting from the union of Q ₂ , P ₂ , T ₂ and C _2. It characterizes all the data associated with the z-test (perceived quality, coding and transmission parameters, terminal parameters, and content parameters):

R ₂ = Q ₂ UP ₂ UT ₂ UC ₂ = (F _{1 z} , .., V _{1 2} ) with t = nq + np + nt + nc Eq. 9

u = union

Table 1: Data constituting the learning set The set of vectors R ₂ , 1 <z ≤ NZ constitutes the learning set (Table 1). A specific procedure is applied to the learning set in order to elaborate the dictionary of representative cases Uk with 1 <k ≤ N. Two cases are possible:

Case 1 (corresponding to the first variant without a vector classification): the number of combinations between the quality levels, the coding and transmission configurations and the characteristics of the terminal and the content is limited (for example NZ <100). In this case, the dictionary UI..N can simply be equal to the training set:

U _UN = (R _k , k = 1 ... NZ). and N = NZ Eq. 10

The limit to the number of combinations can be freely set, for example according to implementation criteria such as the size of the database or the computing power necessary for the optimization module to find the optimal configuration.

Case 2 (corresponding to the second variant with vector classification): the number NZ of combinations R ₂ contained in the training set is very large. An analysis procedure is necessary to generate the N vectors £ Λ..Λ /. of the dictionary that best represent the initial vector set R ₂ . This group of vectors is the one that has the smallest mean distortion with respect to all the vectors of the training set, among the other possible candidate dictionaries. The vectors of this group are then the best representative vectors of the training set, and therefore the relationship between quality and configuration of coding and transmission and the characteristics of the terminal and the content.

Classification algorithms are used. Several authors have proposed solutions for classification into dictionaries. Dynamic Clouds, or LBG Algorithm. The number N of the dictionary vectors is chosen according to the initial number of vectors of the set learning, modeling accuracy and implementation constraints.

The dictionary resulting from the classification procedure constitutes the database DB (FIG. 7). Of course, or could at least use a learning vector that only takes into account perceived quality and coding and transmission parameters. However, it is beneficial to consider the content. The parameters of the terminal need to be taken into account only in the case of a variety of users in the intended application and when it is possible to know the terminal parameter of a given user.

The next step is a search for the encoding and transmission configuration.

The first step has generated a dictionary representative of the relationship between the perceived perceived quality and the coding configuration or the transmission network for certain characteristics of the video content and the data terminal.

The second step uses this dictionary to find a coding and transmission configuration that assures a certain QC target quality at the end-user level. For this, the RECH module looks for this configuration in the DB database (FIG. 4).

The data shown in (Figure 4) is defined below: The vector Q contains the parameters of measured current perceived quality. It is identical to the vector defined by the relation Eq. 5. Q = {yQ _ι , .., VQ _nq ) where nq: number of quality parameters VQi Eq. 11 A date representative of the date of presentation of the video content is also associated with this vector Q. For example nq = 1 Q = quality index between 0 and 100.

The QC vector defines the target perceived quality parameters to achieve. All QC VQQ parameters characterizing the target quality exist in Q, but can of course be of different values. But conversely, all the VO, - Q parameters characterizing the measured quality do not necessarily exist in QC.

For example, the QC vector can be of dimension nqc = 1 and contain a single value gqc corresponding to the target quality to be achieved (for example target quality purchased by the user by contract, passed with the provider of an audiovisual service) for the quality of the audiovisual service

(GQC).

The vector Q must necessarily contain an audiovisual quality value gq obtained by measurement to enable the vector quantization optimization process to work, by comparing gq and gqc. Q can be larger than the dimension nqc of

QC, for example nqc = 1, but nq = 3 in the configuration where Q contains three values Q = (aq, vq, gq) corresponding respectively to the quality obtained by measurement for the audio signal (aq) for the video (vq) and for the audiovisual sector (gq).

QC = ^ QC ₁ , .., VQC _nqc ) where nqc <nq and nqc: number of target quality parameters VQQ Eq. 12

For example nqc = 1 QC = target quality index between 30 and 95.

The vector T contains the characteristic parameters of the terminal. It is identical to the vector defined by the relation Eq. 7.

T - {yτ _x , .., vτ _ιιt ) where nt: number of parameters characteristic of the terminal VT ₁ Eq. 13

For example nt = 1 parameter VT _/ = resolution of the screen.

The vector C defines the parameters of the video content. It is identical to the vector defined by the relation Eq. 8.

C = (vc ₁ , .., VC _nc ) where no: number of parameters of the video content VQ

Eq. 14

For example ne = 1 parameter VCi: activity of a video sequence or type of sequence (slow, fast, average). The vector P defines the desired coding and transmission parameters. It is identical to the vector defined by the relation Eq. 6. P = ψp _x , .., VP _np ) where np: number of coding and transmission parameters VPj Eq. 15

For example np = 1, 2 or 3 VPi, VP ₂ , VP ₃ : transmission power and / or bit rate and / or bandwidth.

The process of searching for the optimal coding and transmission configuration is to extract the P vector giving the coding and transmission pattern to be used to provide the quality level at the user level defined by the quality representative QC vector. under the current constraint conditions represented by the vectors Q, T, and C. The advantage of the vectorization method is that it is not need to measure the perceived quality Q than during the constitution of the dictionary.

The research process is subdivided into three sub-steps: a. Formation of a vector of constraints O. The date associated with the vector Q is associated with the vector of constraints O. This date is representative of the date of presentation of the video content. b. Vector quantization on the vector of constraints O to find the vector U _k of the dictionary corresponding best to the constraint vector O presented as input. vs. Extraction of the vector P of parameters of the coding and transmission system.

Sub-step a) Formation of the vector of constraints O. The vector O representing the set of the current constraints of operation of the system is constituted, in the most powerful case of the union of the vectors T, C, and of a combination Q 'Q and QC vectors. Indeed, each parameter of the vector O must be unique, while the parameters of the vector QC are all present in the vector Q. The final objective is to find the vectors of coding parameters P making it possible to obtain a target quality defined by QC.

Q '= QCu {VQ, I VQ ₁ 3 in QC} with VQ, - defined byg = (Fa, .., Fg, J

Eq. 16 with / = as

For example, in the case where QC is of dimension nqc = 1 and contains a single value gqc corresponding to the target quality to be achieved, and Q is of dimension nq = 3 and contains three corresponding values Q = (aq, vq, gq) respectively for the signal with audio qualities (aq), video

(vq) and audiovisual (gq) obtained by measurement, the vector Q 'resulting from the application of Eq 16 will be Q' = (gqc), corresponding to the audiovisual quality constraint to be obtained from the coding and transmission system.

Then, the vector O is formed by union of T, C, Q '. The resulting vector is of dimension h.

O = Q'vT UC = (VO _ι , .., VO _h ) with h = nq + nt + nc Eq. 17

Sub-step b) Vector Quantification. The vector quantization corresponds to the vector O of parameters VO, - in input, the vector U of the dictionary corresponding to the better to the constraint vector O presented as input. The actual vector quantization is performed on a sub vector S _k of each vector U _k . Indeed, the vector O contains only a subset of the parameters of the vectors U _k . The parameters of U _k not present in O are the P _k parameters of coding and transmission associated with this set of constraints O. Each vector S _k is thus defined by S _k = {V _t / V, 3 in O} with V ₁ defined by U = (V ₁ , .., V ₁ ) Eq. 18 with / = as

The minimization of the distortion between the incident vector O and all the subvectors Sk of the UL.N vectors of the dictionary is made. It makes it possible to identify the vector U corresponding best to the vector of constraints O.

Sub-step c) Extraction of the coding and transmission parameters. The parameters of U not present in O are the coding and transmission parameters P associated with this set of constraints O. It is therefore sufficient to extract from U the vector P representing the coding parameters and which is therefore defined by

P = [V ₁ IV. g in 0} with V _/ defined by U = (V ₁ , .., V ₁ ) Eq. 19 with / = as

The whole operation of the search procedure is illustrated in FIG. 9 for a particular case nq = 4 and nqc = 2:

The parameters of the vector P found, as well as some parameters of the vector U found by vector quantization in substep b if necessary, can then be applied to the rate reduction coding process and the transmission process.

Indeed, some parameters considered as constraint parameters, therefore present in the vector O, can also be useful parameters for defining the transmission configuration.

For example, consider the case where one wishes to optimize the video coding configuration by acting on the two parameters of the spatial resolution and the coding rate. If there are two types of terminals, corresponding to two possible spatial resolutions for the screen, and these terminals are not able to correctly display a coded video of resolution different from that of their screen, the resolution parameter becomes a constraint for the coding method of the rate reduction video. The only parameter of the vector P will therefore be the coding rate. However, the coding resolution (imposed by the terminal) must also be applied to the coding process so that the optimization method is exhaustive. The DB database also has a function of storing the data generated by the perceived quality measurement module, as well as the optimization decisions made by the module RECH. For this purpose, the database DB stores the vectors O and P, represented in FIG. 9, accompanied by the date representative of the date of presentation of the video content which is associated with the vector O.

An alternative to vector quantization is the computation by a law determined logically or empirically in advance, giving the relationship between the perceived quality and the coding configuration or transmission network considered. The optimization procedure f gives the coding and transmission parameters P to be used to obtain a target quality QC, given the characteristics of the terminal T and the video content C ₁ and the current quality level measured Q (FIG. 10). The variables P, QC, T, C, and Q are defined by the equations Eq. 11 to Eq. 15, p. 15. P = f (QC, Q, T, c) Eq. In this case, all the knowledge necessary for the optimization procedure is therefore contained in the deterministic law, located in the module RECH. The DB database does not contain data relating to the optimization procedure.

The optimization approach according to a deterministic law is advantageous because it does not require a database, which can be important. On the other hand, a deterministic law can be easily determined only in the case of a low number of configurations.

The vector quantization approach and representative case database is more advantageous for many configurations.

The invention is particularly applicable to the provision of video sequences on demand from a server by using the vector quantization of the optimal coding rate according to the resolution of the terminal and the quality requested by the end user, depending on the type of sequence desired. This application uses the invention to select the bit rate of video sequences pre-coded and stored on a video server among a number of possible values. The resolution of the user's terminal as well as the desired level of quality are taken into account in order to minimize the throughput required to provide the service, resulting in an optimal use of the transmission network. The transmission network used is, for example, IP (Internet Protocol) DVB (Digital Video Broadcasting) or UMTS (Universal Mobile Telecommunications System).

This application can use an optimization procedure based on vector quantization, as described above.

According to the same notation, this application defines the parameters Q, QC, T, P and C as:

Q = QC = measurement or target of video quality between 0 and 100. The quality measurement method integrates, for example, the method according to PCT patent application WO 2004/047451 mentioned above filed by TDF.

• T = Terminal screen resolution, for example CIF (352 x 288) or QCIF (176 x 144)

• P = Coding rate, in kbit / s. • C = name of the sequence or, in a second variant;

C = type of content (sport, news, ...), to characterize the video content, by type of sequence.

Otherwise, it is possible to characterize the video content by an activity parameter of the image one or more subsequences of a few seconds of a sequence.

Two variants of dictionary construction are presented below, depending on whether the video content is identified by the name of the content, or by the type of content in the dictionary contained in the DB module.

The first variant using the name of the content is described in Figure 11.

1. A number of source video signals are acquired and encoded by rate reduction. The coding is performed according to all the possible resolutions at the terminals, and according to one or more selected bit rates in a range corresponding to the possibilities of the terminals and the transmission network. In the present case, the CIF and QCIF resolutions have been used, and transmission channel rates ranging from 48 kbit / s to 384 kbit / s, for example in 10 kbit / s steps, have been applied in each of these. two cases. 2. Each flow is evaluated by the perceived quality measurement module. The quality Q _z characterizing the coded video sequence is the average quality measured on the sequence.

3. Encoded video streams are stored on a video server. The other data is the dictionary stored in DB: quality Q, transmission channel rate P, terminal resolution T, content name C. It is therefore not necessary to use a classification procedure here, since the size of the dictionary remains modest.

This dictionary can then be used by the RECH module to find the necessary flow, as explained above. A second variant using the type of content instead of the name of the content is described in relation to FIG. 12. The mode of construction of the dictionary is similar: the sequences are coded in all the desired configurations and their quality Q _z evaluated. The difference comes from the use of content type information (eg sport or news) rather than the name. Indeed, the impact of the bit-rate video coding on the perceived quality varies greatly depending on the type of content of the sequence, particularly the presence of additional defects introduced by the transmission channel. For example, sports footage usually requires higher throughput because of more animated content. It is possible to use this property to match a content type to the encoding rate needed for a given quality obtained at reception.

For this purpose, a classification procedure is preferable in order to group the various quality measurements Q ₂ carried out under the same viewing conditions T ₂ and P ₂ coding for several different sequences but of the same type C _z , into a single vector Q, T, P ₁ C. In this embodiment, the classification procedure used is preferably the LBG algorithm with the distance of Eq. 21.

Figure 13 details the implementation of the optimization procedure, during a request submitted by a user. In advance, the user accesses a list of contents stored on a video server, identified by their name and type, for example through an Internet browser; the user selects a desired content and quality level and sends his request. Then, the mechanism using the invention takes place in three stages without user intervention:

1. The user terminal sends to the RECH module its characteristics, the characteristics of the chosen content, the desired quality level QC, and possibly the last quality measure Q in date.

2. The module RECH searches by vector quantization in the database the existing coding rate P for the content C on the video server which provides the QC quality requested, for the resolution imposed by the terminal, and returns this information to the terminal. The parameters received and sent to the terminal are also stored in the database, for example for later analysis. The user terminal accesses the content C selected by the user at the rate P selected by RECH and the user obtains the requested content at a quality QC.

Note that the linear distance between two vectors A and B that can be used here for vector quantization is simpler to implement than the quadratic distance of equation Eq. 4.

^M Figure 14 shows the sequence of operations performed by the module RECH.

RECH receives the characteristics of the terminal T and the content C and possibly the Q quality measurements. It stores these measurements in the database DB via a database management system (DBMS). RECH then performs a search for the best coding or P-transmission configuration from the dictionary also stored in DB. The configuration P is sent to the equipment concerned. In a variant of this first application according to which there is selection of the optimal coding rate according to the resolution of the terminal and the quality required, the method according to the invention is used to minimize the coding bit rate of video sequences by taking only the level of quality to be achieved, resulting in an optimal use of the transmission network. This approach is particularly applicable when the conditions of use of the video service - including the type of terminal - and the content of the service are not very variable. This is for example the case for a video-on-demand service intended to be viewed on television type terminals by the users.

The invention uses the same optimization procedure based on the vector quantization, as described above for said first application, and the same notations. The main difference is that the parameters T and P are empty. The vector quantization is then based on:

. Q = QC = measurement or video quality target between 0 and 100. The quality measurement method integrates, for example, the method according to PCT Application WO / 2004/047451 mentioned above.

. P = Coding rate, in kbit / s.

The same methods of dictionary construction and parameter optimization P, here reduced to the encoding rate, are usable. According to another variant, the method according to the invention makes it possible, for example, to adjust the transmission power as a function of the desired quality and possibly of the type of content, without implementation of a vector quantization.

This application adjusts the transmission power level of the service from a UMTS access network transmitter based on the perceived quality instead of the standard network level parameters used in the network.

UMTS, such as the signal-to-noise ratio Eb / No. The goal is to maintain a given quality level and not a target bit error rate.

Indeed, the sensitivity of a video stream transmitted on a digital network varies according to the type of video content. The presence of motion greatly influences the visibility of the damage caused by transmission errors. In the proposed embodiment, the invention takes advantage of this property to react only when it is necessary to maintain the perceived quality. The invention can use in such a case an optimization procedure based on a deterministic algorithm, as described above.

There is therefore no learning procedure leading to a dictionary.

Based on the same notations as before, this application sets the Q, QC, T, P, and C parameters as follows: • QC = video quality target between 0 and 100. Q = measurement or target of video quality between 0 and 100. The perceived quality measurement method incorporates the method of PCT Patent Application WO 2004/047451 cited above. Q also integrates other measures: the actual rate received by the terminal, and the rate of erroneous data packets received.

• T = (Not used)

• P = Transmitting power (in dB)

• C = Not used or, in a second variant; C = type of content (sport, news, ...)

Figure 13 shows a preferred mode of operation of the application.

The user accesses a list of contents stored on a video server, identified by their name and type, for example through an Internet browser; the user chooses a desired content and quality level. Then, the mechanism using the invention takes place in three stages without user intervention:

1. The terminal periodically sends to the module RECH the last quality measurement Q dated, the desired quality level QC and, in the second variant, the characteristics C of the chosen content.

2. The RECH module applies the optimization procedure from C ₁ QC and Q to find the power P necessary to ensure the quality QC requested under the current perceived quality conditions for the content C. This power P is applied in the network at the broadcast of the video service.

The parameters received from the terminal and then sent to the network are also stored in the database, for example for later analysis.

This optimization procedure which does not implement the vector quantization acts on the power as a function of the perceived quality Q measured. The lower the measured quality Q is from the target quality QC, the more power will vary significantly.

The procedure periodically calculates the new power P, for example every second, from the current power PoId. It can be summarized as follows:

An increase step of the DP power is defined. If IQ - QC | <5 P = PoId

If 5 <I Q - QC I <10 P = PoId - sign (Q - QC) x 1 x DP

If 10 <| Q - QC | <20 P = PoId - sign (Q - QC) x 2 x DP

Otherwise P = PoId - sign (Q - QC) x 4 x DP The sign function (X) returns the sign of X. Thus, the power is increased when Q <QC. For example, DP can represent 1 to 5% of the power.

The method can also be implemented to take into account both the quality measured at the terminal and the type of content, without resorting to vector quantization.

This variant takes advantage of the sensitivity variation of a video stream to transmission errors depending on the type of video content. Indeed, in the case of a transmission over an IP or UMTS network, it is observed that for the same number of lost IP packets, the drop in quality is greater for video sequences having a high-motion content. FIG. 16 shows this phenomenon by taking as a criterion of degradation the proportion of video images lost by transmission: the loss of images is greater for the very animated sequences, which corresponds to a lower quality. This second variant of optimization procedure takes advantage of this property through the following precursor:

Two power increase steps are defined, one for each content type: DPjsport> DP_actualites.

If C = "sport" DP = DP_sport, for example 2 to 10% of the power Otherwise DP = DP_actualites, for example 1 to 5% of the power

If | Q - QC | <5 P = PoId

If | Q - QC | <10 P = PoId - sign (Q - QC) x 1 x DP

If | Q - QC | <20 P = PoId - sign (Q - QC) x 2 x DP

Otherwise P = PoId - sign (Q - QC) x 4 x DP The sign function (X) returns the sign of X. Thus, the power is increased when Q <QC. EXAMPLES

Example 1: For the first variant of the first application (selection of the optimal coding rate according to the resolution of the terminal and the requested quality, using the name of the content)

The table shows a real-world example of a portion of a dictionary used to search for optimal throughput based on a target quality and named content, with a display resolution constraint. The illustrated extract is valid for 5 different encoded contents according to a combination of two resolutions and four rates. These contents are denominated: football ("Foot"), Kayak, Wood, TV Newspaper and Comic Strip (BD).

Table 1: Example of dictionary extract (variant 1)

The coordinates of each line of Table 1, that is to say of each vector of the dictionary, can be related to the definitions of vectors Q, C, T and P previously defined at Eq. 5, 6, 7, and 8 as follows:

Q = (pqos) and nq = 1 pqos = quality of the sequence between 1 and 100

C = (sequence name) and ne = 1

T = (image size) and nt = 1

P = (bit rate) and np = 1

The application of the vector quantization procedure with a QC vector containing a target quality value of pqos then makes it possible to select the optimum value of the "bit rate" parameter. In this case, the distance between two coordinates "image size" (or "bit rate") is zero if the two coordinates of the same vector are equal, otherwise it can be chosen for example equal to 100 so as to be of an order of magnitude comparable to the coordinate pqos.

Then, as indicated above in the description (substep c), the encoding configuration sent to the encoder is composed of the vector P and possibly some elements of the vector U: in this case, the parameters "image size" or "bit rate" is this configuration.

Example II: For the second variant of the first application (selection of the optimal coding rate according to the resolution of the terminal and the requested quality, using the type of content)

The table shows a real-world example of a dictionary used to find the optimal throughput based on a target quality and a content type (news or "News" and "Sport"), with a display resolution constraint.

Table 2: Dictionary Example (Variant 2)

The coordinates of each line of Table 2, that is to say of each vector of the dictionary, can be related to the definitions of vectors Q, C, T and P previously defined at Eq. 5, 6, 7, and 8 as follows:

Q = (pqos) and nq = 1

C = (content type) and ne = 1 T = (image size) and nt = 1 P = {bit rate) and np = 1 The application of the vector quantization procedure with a QC vector containing a value The target quality of pqos then makes it possible to select the optimal value of the parameter "bit rate". In the present case, the distance between two coordinates image size or bit rate is zero if the two coordinates are equal, otherwise it can be chosen for example equal to 100 so as to be of an order of magnitude comparable to the coordinate pqos.

Then, as indicated above in the description (substep c), the coding configuration sent to the encoder is composed of the vector P and possibly some elements of the vector U: in this case, the parameters "bit rate" and "image size" constitute this configuration.

Claims

A method for transmitting a variable bit-rate audio and / or video program over a transmission channel, characterized in that it implements an adjustment of at least one coding and / or transmission parameter. function of at least one setpoint vector with at least one dimension representing a reception quality desired by said end user.

2. Method according to claim 1, characterized in that it is implemented for the provision of video sequences to a user from a server and in that it implements the choice by the user of a video sequence and a quality level chosen QC.

3. Method according to one of claims 1 or 2, characterized in that a said transmission parameter is the bit rate and / or the type of modulation and / or the transmission power.

4. Method according to one of claims 1 to 3, characterized in that said adjustment is made from a deterministic relationship between the desired reception quality and the coding parameter and / or transmission.

5. Method according to one of claims 1 to 3, characterized in that said adjustment is implemented as a function of a distance between said target vector and a measurement vector representing said quality of reception measured at said end user.

6. Method according to claim 5, characterized in that the quality of reception is measured on a fixed duration sequence of said program.

7. Method according to one of claims 5 or 6, characterized in that said adjustment is performed by modifying the transmission power P as a function of a distance between the target vector and the measurement vector.

8. Method according to one of the preceding claims, characterized in that said adjustment is also performed as a function of at least one parameter of the program content.

9. Method according to one of the preceding claims, characterized in that a parameter of the content is an activity parameter and / or a parameter assigned to the name of the program and / or the type of program.

10. Method according to one of the preceding claims, characterized in that said adjustment is also performed as a function of a characteristic parameter of the terminal.

11. The method of claim 10, characterized in that a characteristic parameter of the terminal is the resolution of an image displayed on said terminal and / or the bandwidth.

12. Method according to one of the preceding claims, characterized in that it implements the elaboration of a dictionary from a learning set comprising NZ vectors R characterizing the data of NZ tests, each vector R _z ( Z varying from 1 to NZ) of a rank test z, resulting from the union of a vector Qz representing the perceived quality of this rank test z, a vector Pz representing the coding parameter (s) and / or or transmitting this test of rank z, and possibly a vector Tz representing the parameter or parameters of the terminal of this rank test z, and / or a vector C _z representing the content parameter or parameters.

13. The method of claim 12, characterized in that the dictionary is constituted by the vectors of the training set.

14. Method according to claim 12, characterized in that the dictionary is obtained by a vector classification algorithm from said learning set, and is constituted by a group of N vectors (with N <NZ) having a distortion minimum average compared to the NZ vectors of the learning set.

15. Method according to one of claims 13 or 14, characterized in that the adjustment is carried out by determining by vector quantization the dictionary vector corresponding best to a constraint vector representing at least the desired quality.

16. The method of claim 15, characterized in that the constraint vector consists of the union of a vector representing the desired quality and a vector representing at least one content parameter and / or a vector representing at least one parameter of the terminal.

17. Method according to one of claims 1 or 2, characterized in that a said adjustment for example of the transmission power P is achieved by step DP according to the difference between the perceived quality measured Q and the target quality QC for the video program.

18. Process according to claim 17, characterized in that:

- If | Q - Qc | is less than a first threshold, the power P is not modified

- If | Q - Qc | is between the first threshold and a second threshold, the power is increased or decreased by the step DP according to whether the sign of Q - QC is respectively negative or positive

- If | Q - Qc | is between the second threshold and a third threshold greater than the second threshold, the power is increased or decreased by kDP with 1 <k = 2 depending on whether the sign of Q - Q _c is respectively negative or positive.

19. The method of claim 18, characterized in that the DP step is variable depending on a type of content associated with the video program.