KR20010041441A - Method and device for encoding a video signal - Google Patents

Abstract

In video communications, the quality and delay of transmission rely on a bit rate control strategy that allows the bandwidth of the number of bits generated for a given setting of the encoder. A pre-analysis step based on the approximate prediction (by the use of empirical rules for motion information and statistics on the content of the picture) to give the encoder appropriate guidelines to consequently predict the generated bits of the number. A determination step is provided, followed by a decision step provided to adjust the quantization step size and the speed of successive pictures.

Description

Method and device for encoding a video signal

In low bit rate applications (television telephony, television conferencing), a system for video compression as suggested in the H.263 standard is recommended. Video encoders according to this standard are described, for example, in B.Furht et al., Kluwer Academic Publishers, 1997 edition, "Operation Decision Algorithms for Video Compression," chapters 2-30-35. The H.263 video encoder as shown in FIG. 1 is based on motion compensated prediction from the previous picture to the current picture, followed by an orthogonal transform (such as DCT), which does not associate the picture elements, transforms and quantizes them. Some low order coefficients for encoding the predicted error and the quantization and encoding operations concentrate the energy to reduce the spatial redundancy of the pictures. At least the first picture is for reference and is encoded without temporal prediction (ie according to "internal mode" or I mode), and sometimes at least one picture in every n pictures is also coded according to the internal mode. Another form of picture, P, corresponds to P pictures, ie pictures that are temporarily predicted from previous pictures.

The present invention relates to a method for encoding successive pictures of a video signal, comprising dividing each successive picture into a plurality of sub-pictures, and converting each sub-picture into coefficients. Quantizing the coefficients to an applied step size, coding the quantized coefficients, and controlling the step size according to a target value for the number of bits for encoding each successive picture. Include. The invention also relates to a corresponding device. The invention is particularly adapted to real time communication at low bit rates in accordance with so-called H.263 recommendations.

1 is a basic video compression schematic in accordance with the H.263 standard.

2 is a general structure of an encoder according to the invention.

3 shows how the schematic diagram of FIG. 2 works.

As shown in Fig. 1, the consecutive pictures P arrive at the input of the encoding device (input video) regularly. A coding branch comprising a discrete cosine transform circuit 12, a quantization circuit 13 and a variable length encoder 14 (DCT, Q, VLC, respectively) processes the images (the circuit 12 In effect receives the difference between the input pictures available at the output of a sub-ctor 25 and the predicted ones, and sends the obtained coded variable bit rate bitstream to a buffer 15, the output of which is Is the output constant bit rate bitstream of the H.263 video encoder. The output bitstream is also sent to the bit rate control circuit 30 for buffer adjustment. A prediction branch is provided, an inverse quantization circuit 21 (Q- ¹ ), an inverse DCT transform circuit 22 (DCT- ¹ ), and an adder 23 (which carries the reconstructed previous picture RPP). And a temporal prediction circuit 24 (transfers the predicted picture PP sent to the subtractor 15 and the motion vectors MV sent to the variable length encoder 14, wherein the prediction is carried out with a picture skipping circuit). Based on a block-matching search performed between the current picture CP available at the output of 26) and the reconstructed one RPP available at the output of the adder), and the subtractor 25 It includes continuously. The output of the temporal prediction circuit 24 is also returned to another input of the adder 23 in view of the reconstruction of the previous picture RPP used for the temporal prediction.

The H.263 standard defines a hierarchical bit stream sentence structure with four layers of the hierarchy, the picture level and group level of blocks (GOB), macroblock level (MB) and block level (8x8 picture elements). Or pixels) and the block is the basic unit on which the DCT operates. The macro block includes four luminance blocks (covering 16 × 16 areas in the image) and two chrominance blocks. The operation determination and compensation performed at the reconstructed previous picture RPP in the circuitry 24 of the prediction branch operate on macro blocks. The feedback connection 31 between the buffer 15 and the quantization circuit 13 allows for obtaining finer or coarser quantization. The coarseness of the quantization is defined by a fixed quantization matrix that sets the quantization parameter for the first three layers (blocks, macroblocks, GOBs) and the relative coarseness of the quantization for each DCT coefficient. . The picture skipping circuit 26 provided at the input of the encoder is also used in a possible way to reduce the bit rate (while maintaining a receivable picture quality). The number of skipped pictures is variable and depends on the output buffer fullness, and the feedback connection 31 provided for buffer adjustment is therefore related to image skipping (and intra / inter selection as well as quantization step size changes). Which is controlled by the circuit 41 which activates the first switch 42 and the second switch 43).

The feedback connection allows to give guidelines to the encoder. The problem of bit rate control is actually how to determine which encoder setting should be selected, given a given bit rate and an input picture? It can be formulated as The pictures have a constant picture rate approach and are periodically taken accordingly and the quality is adapted to the complexity of each successive picture (to maintain the target bit rate) and so the variability from one picture to another becomes large, Only when the constant quality approach and therefore the encoder has finished processing the previous picture, ie at high variable picture rates adapted to the complexity of the pictures, it will be possible for the pictures to be processed in a fixed quantization step.

It is therefore an object of the present invention to present an improved encoding method in which a compromise between picture speed and quality is investigated, in order to ensure that the delay between the input pictures and the displayed ones is as regular as possible to recover the scene. Consider the fact that it should be controlled (as constant as possible).

For this purpose the present invention relates to a method as described in the preamble of the above description, wherein the control step comprises a pre-analysis sub-step, wherein the operation information between the previous and current pictures is displayed. Is based on the determination of the number of bits each used to code and to code the coefficients, and the decision sub-step provided to adjust the quantization step size and the speed of the successive pictures.

Certain aspects of the invention will now be described with reference to the embodiments described below and will be considered in connection with the accompanying drawings.

In the coding chain as shown in FIG. 1, the starting point of the feedback control performed by the output buffer 15 is to match the number of bits to be used to code each successive current picture with the number of bits available in the transmission channel. will be. When studying the bit generation during the coding operation of each picture, it appears that the largest amount of bits to be transmitted relates to the content of the picture. In fact, bits are used to code information at the picture level, GOB level, macroblock level and block levels, but the most expensive part by the number of bits generated is the macroblock and block levels, which is the operation decision and DCT It is related to coefficient coding and depends entirely on the complexity of the current picture.

The quality of communication in such a coding chain is variable of the generated bits and depends on the part that depends on the quantifier. According to the invention, a pre-analysis of the current image is provided which roughly predicts how many bits will be generated according to the settings of the encoder. The pre-analysis described in more detail below allows for predicting the number of bits generated for each possible quantization step (pre-analysis step) of the encoder. A decision step follows, where a setting for the encoder is found after comparing the bits of the number with the required one. If the corresponding quantization step is according to a previously set quality region, the picture is coded with it (this means that transmission is not allowed when the quality is too bad, corresponding to too large quantifier step sizes). . Otherwise, in order to meet the bandwidth requirements, the worst allowed quantization step is selected first and the reduced picture rate is calculated (and selected according to the picture skipping circuit 26). The calculated settings are then used for the encoding process of the picture. After the encoding operation of each GOB, the bit rate control allowed by the feedback connection for buffer adjustment checks if a deviation between the predicted and required numbers of bits appeared during the encoding operation, and if necessary The setting of the encoder is modified by modifying the quantization step between two consecutive GOBs (permitted changes are only prus or minus one).

The generated bits can be divided into two parts, the first one corresponding to the headers and the second one corresponding to the actual content of the current picture. The calculation of the first part is easy, but the calculation of the second part is more complicated.

In order to transfer the contents of the picture, two kinds of information are certainly needed: (a) information of the operation between the current picture and the previous picture and (b) the current macro block and the corresponding part of the previous picture. Are chrominance and luminance changes.

The first pre-analysis sub-step related to the motion information is based on an approximate prediction of the number of bits needed to code the motion vectors. In more detail, it has been found that a rule of thumb can be established that connects the average operation of the complete picture with the number of bits required to code all the motion data. The law may be expressed in the form of the following formula (1).

PNB = 4.N.log (1000.Mean_mv / N) (1)

here:

N = number of macroblocks with nonzero motion vectors,

Mean_mv = (sum of motion vectors) / (number of macro blocks per image),

PNB = a determined number of bits (relative to the average operation of the entire picture) used to code the motion information of the picture.

The law has the advantage of being simple and fast.

The second pre-analysis sub-step related to the number of bits for coding the DCT coefficients is based on the prediction made at the macro block level. For each quantization step, the sum of the absolute differences between the luminance data of the current and the reference macro blocks (= SAD, which is the correlation between the original macro block of the current picture and the displaced macro block of the previous reconstructed picture). Is a relationship measure, and according to equation (2),

(2)

Statistics of the number of bits used were made against x, y = displacement coordinates and N = size of the block, generally 8 or 16), indicating that the number of bits does not have a linear change with respect to the SAD. This can be because of the coding mode of the DCT coefficients in the H.263 standard, which is a variable length coding mode for more frequent coefficients and a fixed length coding mode for others: This is why for each quantization step It can be explained whether three different areas, one for small SAD and another for medium and large SAD, are observed.

This results in different predictions depending on the SAD region: A tradeoff was made for the three regions working in the full range (1 to 31) of the quantization step. Third order polynomial approximation laws were calculated for each quantization step for SAD <500, 500 <SAD <1000, 1000 <SAD <1500, and a fixed value was chosen for SAD> 1500. The tradeoff between accuracy and computational complexity seems rather good.

In practice, at the beginning of the coding, operation determination of the entire image is made. The SAD of each macro block is known and the predicted number of bits is then calculated for each quantization step to determine which quantization step gives the closest prediction to the desired number of bits. If the calculated quantization step is too high in terms of minimum quality, the quantization step is set to the maximum value allowed. Knowing the number of bits sent on the line, the time to capture the next picture is then calculated.

In the encoder according to the invention shown in FIG. 2, the pre-analysis sub-steps referred to as a single reference 201 are followed by a decision sub-step 202. The reference " coder 203 " designates the association of all elements of FIG. 1, except for the buffer 15 and the bit rate control circuit 30.

For the basic schematic of FIG. 1, FIG. 2 illustrates at what level the preanalysis works in the coding chain, and FIG. 3 shows how the schematic of FIG. 2 (ie, bit rate control according to the present invention) works. . The adjustment is performed here by executing a set of software instructions that control the calculation steps and the test steps or similar steps as now described.

The first step 31 takes into account the input picture and the bandwidth BW according to relation (3) of the following format, the fullness of the output buffer FOB and the allowed target number of bits for the target frame rate TFR. It is provided to calculate T _b .

(3)

The second step comprises a predicted number of bits P _n useful for coding the actual picture with a smaller quantizer (or quantization) step that appears to match the bandwidth, the quality and the frame rate. A calculation operation 321 provided for calculation. The numbers T _b and P _n are then compared. (Test operation 322): If P _n is greater than T _b (output Y), the coding step of FIG. 2 will be performed in conjunction with the quantization step. On the other hand, if P _n is less than T _b (output N), 1 is added to the quantization step and the operation 321 is repeated, and the test operation 322 is repeated with the modified value of P _n . .

When P _n is greater than T _b , a quality test 33 is performed. If the quantization device is below a predetermined quality threshold Q _n-1 &lt; Q _max , the frame rate FR is determined by the target frame rate (connection). (331)), on the other hand, if not the case, a new smaller frame rate is calculated according to the predicted number of bits for the minimum permitted quality (connection 332). A coding step 341 is then performed to code each group of blocks GOB.

If the coding prediction is according to the actual coding (sub-step 343) (test prediction OK?), Then the last step is for each new GOB (test new GOB? 342). It is provided for checking. If YES, the next coding step will continue with the same quantizer (connection 344), otherwise add or subtract 1 from the computed quantizer in terms of reducing the drift (sub-step 345). , New amount = +/- 1).

Claims (2)

In a method of coding a video signal of successive pictures, Dividing each successive picture into a plurality of sub-pictures, Converting each sub-picture into coefficients, Quantizing the coefficients to an applied step size; Coding the quantized coefficients; Controlling said step size in accordance with a target value for the number of bits for encoding each successive picture, said step of determining the number of bits each used to code motion information between previous and current pictures and to code said coefficients And a determining sub-step based on the pre-analysis sub-step and a decision sub-step provided for adjusting the quantizing step size and the speed of the successive pictures. An apparatus for encoding a video signal of successive pictures, Means for dividing each successive picture into a plurality of sub-pictures, An encoder for continuously encoding the sub-pictures or a group of sub-pictures, comprising: an image converter for converting each sub-picture into coefficients and a quantizer for quantizing the coefficients to an applied step size With the encoder, Control means for controlling the quantization step size according to a target value for the number of bits for encoding the applied picture, The apparatus comprises a pre-analysis stage provided for coding motion information between previous and current pictures and for determining the number of bits each used to code the coefficients, and for adjusting the quantization step size and the speed of successive pictures. And a provided decision stage.