CN1738424A - A rate control method for fine-grained scalable coding based on H.264 - Google Patents

Abstract

The invention relates to encoding rate control method of fine particle flexible encoding based on the H.264/AVC. The invention combines the basic flow and the reinforcement flow at the encoder into one code flow document, so only one code flow document and less information of reinforcement encoding should be transmitted; uses a encoding rate control method of FGS based on H.264/AVC to self-adaptive select the QR (quantizing parameter) according to the result of optimizing the distortion; operates the encoding rate control of GOP (picture group) grade on the basic layer in the encoder; operates the encoding rate control on the reinforcement successively fined. Compared to other encoding rate control methods based on the multilayer video encoding, the invention has higher video quality and smooth change of PSNR (peak signal-to-noise ratio). In addition, the invention can use terminal which is arranged with different temporal resolution (frame speed) and different spatial resolution (image mode) to realize the real-time interception and decoding limited in a certain target bit rate at the receiving end. The result of testing said method indicates: the mean brightness peak signal-to-noise ratio (Y-PSNR) in the GIF mode is better than the one in JM8.6+FGS mode (using the basic layer encoding rate control of JM8.6 and using the bit plane truncation method of FGS as the reinforcement encoding rate control) to reach the 2.45 dB, while the former more matches the target bit rate.

Description

A rate control method for fine-grained scalable coding based on H.264

technical field

The present invention relates to a video coding method, in particular to a code rate control method based on H.264/AVC fine granular scalable coding.

Background technique

Rate control is the key technology of H.264/AVC video codec. It is imperative for the encoder to allocate the appropriate number of bits for each video frame and to make the variable bitrate as smooth as possible. There are two different rate control schemes for H.264/AVC: one is TM5 derived from MPEG-2, which is not suitable for low bit rate video applications; the other is based on traffic model, linear model and rate-distortion optimization (RDO ) control strategy.

Although H.264/AVC has high encoding efficiency, its various prediction modes and RDO options bring new problems to the rate control. Pornthep Navakitkanok and others proposed a rate control scheme that can effectively allocate the number of bits and monitor the fullness of the buffer, but only considers the case of low latency. Minqiang Jiang et al. apply the MAD (mean absolute difference) of the previously encoded frame to predict the complexity of the current frame, and use the MAD ratio to adjust the target bit estimate. But MAD ratio is obtained by linear prediction, which is inaccurate when the scene changes in the sequence; moreover, it only reduces the change of peak signal-to-noise ratio PSNR, and it is difficult to improve the average PSNR.

Some literatures apply linear models to predict the MAD of the current frame. The problem with this approach is that the target bits only depend on the fullness of the buffer without taking into account the content of the video frame. If the buffer is nearly full, then only a few target bits are allocated for new encoded frames. This will result in a non-uniform distortion of the video sequence. Moreover, the MAD rate control algorithm can only express the similarity between the current frame and its matching reference frame, and it is difficult to better express the motion content of the video frame. Some literatures propose a rate-quantization (R-Q) model, which approximates the relationship between the number of coded bits of a video frame and the quantization step at the frame level, but it is only suitable for video sequences with less motion. Chi-Wah Wong et al proposed a rate control method based on a linear bit rate and a linear distortion model, using the probability of the non-zero coefficient of the prediction error of the quantized macroblock, but the probability factor is only determined by the average QP of the previous frame (quantization parameter) obtained, not accurate enough. Hongtao Yu et al. use the motion complexity obtained from the actual number of bits of the previously encoded frame to represent the complexity of the motion content of a frame. But this is only a rate control method at a constant bit rate.

All the above rate control methods are designed for H.264/AVC single-layer video coding. Bit rate changes in a short period of time can be resolved by receiver buffering. However, when the code rate changes for a long period of time, if the buffer method is still used, it will be difficult to be effective due to the limited buffer capacity. The code rate variation problem over a long period of time should be solved by a scalable coding method based on SNR (quality), time, or space.

Contents of the invention

The object of the present invention is to provide a kind of code rate control method based on H.264/AVC fine grain scalable coding, compared with other multilayer code rate control methods, its average luminance peak signal-to-noise ratio (Y -PSNR) has an average gain of 2.45dB compared to JM8.6+FGS (using JM8.6's base layer code rate control and FGS's bit plane truncation method for enhancement layer code rate control), and is closer to the target bit rate match.

To achieve the above object, design of the present invention is:

For any video sequence of CIF or QCIF, in addition to using an appropriate size of QP value to generate the base layer (BL) to ensure that the minimum bit rate constraint is met, another set of FGS scalable layers is added, and the bit rate can be higher than the target bit rate . Then use the fine-grained SNR scalable process to reduce the final code rate to meet the bit rate constraints.

The steps of the above FGS are as follows: first use the H.264/AVC intra-frame or inter-frame residual coding to obtain the minimum SNR quality, and then continuously refine the coding of the transform coefficients by continuously reducing the quantization step size, and at the same time apply a similar "sub- A modified CABAC (Context-Based Adaptive Binary Arithmetic Coding method) of Planar Coding method to achieve fine-grained SNR scalable process. In such successive refinement coding mode, the coding process of each enhancement layer for transform coefficient refinement level is divided into 3 scans: coarse scan, refinement scan and residual scan. After each scan, the corresponding transform coefficients are transmitted.

The successively refined FGS of the present invention can truncate the corresponding NAL (Network Adaptation Layer) unit at any point of any enhancement layer. For any spatial (CIF/QCIF video format)-temporal (30Hz/15Hz frame rate) resolution, the minimum bit rate is the bit rate that the corresponding base layer must transmit at which the reconstructed video is acceptable Minimum video quality. Above this code rate, as long as the channel bandwidth and terminal configuration allow, the successive refinement NAL units of the corresponding space-time layer can be intercepted to extract code streams of any code rate.

According to above-mentioned design, technical scheme of the present invention is:

A code rate control method based on H.264/AVC fine-grained scalable coding, characterized in that an encoder end combines two code streams of a basic layer stream and an enhanced layer stream, and adopts a fine-grained code stream based on H.264/AVC The code rate control method of scalable coding generates a code stream file at the receiving end, and realizes that only one code stream file and a very small amount of enhancement layer coding information need to be transmitted; the steps are:

(1) Fine-grained scalable coding based on H.264/AVC: adaptively select quantization parameters according to the result of rate distortion, and encode the total bit stream;

(2) Base layer flow rate control: perform group-of-picture rate control on the base layer at the encoder end;

(3) Enhancement layer code rate control: perform successively refined code rate control on the enhancement layer at the encoder end;

(4) Code rate control performed when the target bit stream is extracted from the total bit stream at the decoding end, and a code stream of any code rate is extracted.

The above-mentioned fine-grained scalable encoding based on H.264/AVC requires a motion compensation time decomposition process for each spatial layer to provide temporal scalability; the specific steps are as follows:

(1) Input the original sequence, that is, the full precision time resolution sequence;

(2) time decomposition predictive filtering, obtains high-pass signal, calculates its quantization parameter QP _H ; All the high-pass signals of this layer form a time enhancement layer; Time decomposition update filtering, obtains low-pass signal, calculates its quantization parameter _QPL ;

(3) Repeat step (2) from the highest layer to the last layer (lowest layer) by reducing the 1/2 precision time resolution each time, to obtain all low-pass signals of the last layer to form the time basic layer;

(4) Residual texture information undergoes residual coding, and motion description information undergoes motion coding to obtain SNR scalable successive refinement quality layers.

The specific steps of the above-mentioned basic layer code rate control are as follows:

(1) Define the complexity measure of the image group by calculating the average space complexity;

(2) Initially set the weighting factor ρ=0 between the buffer change and the complexity requirement, the initial bit budget of the sequence is Br, the initial buffer fullness is β ₁ , and the cycle starts from the first image group;

(3) Calculating the gth image group allocation bit number Bt(g);

(4) Use the tentatively allocated bit budget Bt(g) to detect the buffer state;

If the filling degree of the buffer zone is less than 80%, go to step (5), otherwise adjust ρ to increase it by 0.1, and repeat step (3);

(5) Update the buffer fullness B _g , and update the sequence remaining bit budget Br;

(6) Until the last GOP, the base layer GOP-level rate control ends.

The above enhancement layer code rate control extracts the required bit rate code stream from the total code stream, and the specific steps are as follows:

(1) Set the required target bit rate Rt;

(2) Obtain the base layer bit rate R ₀ from the base layer code rate control process; when R ₀ > Rt, extracting the target bit stream fails and exits; otherwise proceed to the next step (3);

(3) The target bit rate is revised to Rt=Rt-R ₀ ;

(4) Calculate the bit rate R _F of a successive refinement layer;

(5) If R _F ≤ Rt, then modify the target bit rate to Rt=Rt-R _F , repeat step (4);

Otherwise, the refinement packet is truncated, and Rt=0, the target bit stream is successfully extracted, and the rate control of the enhancement layer ends.

The above-mentioned technical scheme is given further detailed description below:

Fig. 1 shows following three steps in the general technical scheme of the present invention:

(1) FGS encoding based on H.264/AVC:

精度时间分辨率序列；再上一级时间分解阶段的结果是第二个时间增强层，……。在时间分解预测滤波的过程中，给定信号的奇数点样本是由偶数点样本的线性组合来预测的，新的预测值和原奇数点样本之差形成一个高通信号。在更新滤波过程中，预测残差(高通信号)的线性组合与输入信号的偶数点样本之和构成低通信号。量化参数QP_H(k，i)和QP_L(k，i)分别用于编码高通信号和低通信号，而这些高通信号H_k[i]和低通信号L_k[i]是原始输入序列的第i帧的FGS的第k(k＝0，...，n-1)个时间分解层所获得的。其中n是图像组中应用的总的时间分解层数。然后，时间可伸缩过程输出的残差纹理信息经过残差编码，运动描述信息经过运动编码，得到SNR可伸缩逐次精细化质量层。即：从较低空间层的运动信息出发，来作较高层的运动预测；从较低空间层的纹理信息出发，做较高空间层的空间预测来消除冗余。As shown in Figure 2, a motion-compensated temporal decomposition process is performed for each spatial layer, thereby providing temporal scalability. The original sequence is a full precision time resolution sequence. The result of the lowest time decomposition stage is the time basic layer, and the result of the upper level time decomposition stage is the first time enhancement layer. These two layers together form

Precision temporal resolution sequence; the result of the next higher temporal decomposition stage is a second temporal enhancement layer,  … . In the process of time-decomposition predictive filtering, odd-numbered samples of a given signal are predicted by a linear combination of even-numbered samples, and the difference between the new predicted value and the original odd-numbered samples forms a high-pass signal. In the update filtering process, the linear combination of the prediction residual (high-pass signal) and the sum of the even-numbered samples of the input signal form a low-pass signal. Quantization parameters QP _H (k, i) and QP _L (k, i) are used to encode high-pass signals and low-pass signals, respectively, and these high-pass signals H _k [i] and low-pass signals L _k [i] are the original obtained by the kth (k=0,...,n-1) time-decomposition layer of the FGS of the i-th frame of the input sequence. where n is the total number of temporal decomposition layers applied in the image group. Then, the residual texture information output by the temporal scalability process is residual-coded, and the motion description information is motion-coded to obtain SNR scalable successive refinement quality layers. That is: starting from the motion information of the lower spatial layer, the motion prediction of the higher layer is performed; starting from the texture information of the lower spatial layer, the spatial prediction of the higher spatial layer is performed to eliminate redundancy.

(2) Basic layer code rate control:

As shown in FIG. 3 , the rate control of the basic layer in this method is rate control at the GOP level. To assign the number of bits to each GOP, first define a complexity measure for each GOP by computing the average space complexity as follows:

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; MV</span>}}_{\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; /</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; MV</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ,</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

where Cg _,i is the complexity measure of the i-th frame in the g-th GOP, Hg _,i is the number of bits in frame (g,i) representing non-texture information (e.g. frame header/syntax, motion vector), R _{g, i} is the total number of bits of the coded frame (g, i), R _{g, i} -H _{g, i} is the budgeted number of bits for coding the texture information of the coded frame (g, i), R(g)-H (g) is the average number of bits for encoding texture information of the g-th GOP. MV _g,i is the number of bits of the motion vector of frame (g,i), and MV(g) is the average number of bits of the motion vector of the g-th GOP group. N(g) represents the frame number of the g-th GOP. Let ρ represent the weighting factor between buffer variation and complexity requirements. ρ=0 means that the bit allocation scheme completely obeys the frame complexity; ρ=1.0 means that the bit budget is allocated evenly without considering the frame complexity; 0<ρ<1.0 means that the bit allocation is a trade-off between the buffer and quality constraints. In addition, let R be the channel code rate, F be the selected frame rate, and β _max be the maximum buffer fullness. The following GOP-level round-robin bit allocation is for the associated N GOPs.

(A) Initially set ρ=0, the bit budget is B _r , and the initial buffer fullness is β _l . Start looping from the first GOP.

(B) assign the number of bits to the gth GOP according to the formula:

${\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; \&Center\; Dot;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&rho;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\frac{\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span>}{\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; g</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; B</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; .</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

R/F represents the average target number of bits per frame.

(C) Use the heuristically allocated bit budget B _t (g) to detect the buffer state: if

(4)

Then go to step (D): otherwise, adjust ρ to increase it by 0.1, and then go back to step (B).

(D) Update buffer status:

(5)

And update the remaining bit budget B _r =B _r -B _t (g). Then go to step (B) to do the calculation of the (g+1)th GOP.

GOP-level bit allocation allocates a bit budget for each GOP while satisfying buffer and quality constraints. The bit allocation scheme at the frame level is similar to the bit allocation at the GOP level, and it also allocates the bit budget according to the frame complexity.

(3) Enhancement layer code rate control

The code stream of the required bit rate is extracted from the total code stream, and the following method is adopted. As shown in Fig. 4, let R _t denote the target bit rate of a certain space-time resolution S _t -T _t , and let R ₀ denote the base layer bit rate under this space-time resolution. If R ₀ >R _t , then the required space-time-code rate point (such as 176×144@15fps: 50kbit/s means extracting a 50kbit/s code stream when the QCIF format frame rate is 15fps) cannot be obtained from the total code stream Extracted from; otherwise:

(A) The target bit rate is modified to $R_t\&DoubleLeftArrow;R_t-R_0.$

(B) From the lowest supported spatial resolution to the target spatial resolution, successive refinement encoding packets need to be calculated. For each spatial resolution, successive refinement packets are processed in the order from the lowest refinement coding layer to the highest refinement layer. For each successive refinement layer of the space-time resolution ST _t , the following processing is performed:

(a) Let R _F denote the bit rate of the i-th successive refinement of the spatio-temporal resolution ST _t .

(b) If R _F ≤ R _t , then the corresponding successive refinement packets are all included in the extracted bit stream, and the target bit rate is modified as $R_t\&DoubleLeftArrow;R_t-R_f.$

(c) Otherwise, the corresponding successive refinement packet is truncated, and the target code rate is reset to zero: R _t =0. Let L be the original size of the successive refinement packet, then after truncation, the length becomes

Compared with the prior art, the present invention has the following obvious substantive outstanding features and significant advantages: usually for the FGS coding method, the base layer stream and the enhancement layer stream are coded into two video streams and transmitted separately; and the present invention Combine these two code streams at the encoder side to generate a code stream file, and only need to transmit this code stream and a very small amount of enhancement layer coding information. Adopt a code rate control method based on FGS of H.264/AVC, adaptively select QP (quantization parameter) according to the result of rate-distortion optimization, and perform GOP (group of picture) level code rate control on the base layer at the encoder end , while performing refined bit rate control on the enhancement layer, compared with other bit rate control methods for multi-layer video coding, the video quality is higher, and the PSNR (peak signal-to-noise ratio) changes more smoothly. At the receiving end, under a certain target bit rate constraint, it can be intercepted and decoded in real time by terminals configured with different temporal resolutions (frame rate) and different spatial resolutions (image format). Tests show that the average luminance peak signal-to-noise ratio (Y-PSNR) of the present invention is 2.45dB better than the JM8.6+FGS method in the CIF format, and more closely matches the target bit rate.

Description of drawings

FIG. 1 is a block diagram of the code rate control procedure of H.264/AVC-based fine-grained scalable coding according to the present invention.

FIG. 2 is a program block diagram of H.264/AVC-based fine-grained scalable coding in FIG. 1 .

FIG. 3 is a block diagram of the basic layer code rate control program in FIG. 1 .

FIG. 4 is a block diagram of the code rate control program of the enhancement layer in FIG. 1 .

Detailed ways

An embodiment of the invention is:

Referring to Fig. 1, the control method of the fine-grained scalable coding based on H.264/AVC is to combine the two code streams of the basic layer stream and the enhanced layer stream at the encoder end, and adopt the fine-grained scalable coding based on H.264/AVC The code rate control method of scalable coding generates a code stream file at the receiving end, and realizes that only one code stream file and a very small amount of enhancement layer coding information need to be transmitted; the steps are:

(1) Fine-grained scalable coding based on H.264/AVC: adaptively select quantization parameters according to the result of rate distortion, and encode the total bit stream;

(2) Base layer flow rate control: perform group-of-picture rate control on the base layer at the encoder end;

(3) Enhancement layer code rate control: perform successively refined code rate control on the enhancement layer at the encoder end;

(4) Code rate control performed when the target bit stream is extracted from the total bit stream at the decoding end, and a code stream of any code rate is extracted.

Referring to Fig. 2, the code rate control method of the above-mentioned fine-grained scalable coding based on H.264/AVC is characterized in that the fine-grained scalable coding based on H.264/AVC requires A motion-compensated time-decomposition process is performed to provide temporal scalability; the specific steps are as follows:

(1) Input the original sequence, that is, the full precision time resolution sequence;

(2) time decomposition predictive filtering, obtains high-pass signal, calculates its quantization parameter QP _H ; All the high-pass signals of this layer form a time enhancement layer; Time decomposition update filtering, obtains low-pass signal, calculates its quantization parameter _QPL ;

(3) Repeat step (2) from the highest layer to the last layer (lowest layer) by reducing the 1/2 precision time resolution each time, to obtain all low-pass signals of the last layer to form the time basic layer;

(4) Residual texture information undergoes residual coding, and motion description information undergoes motion coding to obtain SNR scalable successive refinement quality layers.

Referring to Figure 3, the specific steps of the above-mentioned basic layer code rate control are as follows:

(1) Define the complexity measure of the image group by calculating the average space complexity;

(2) Initially set the weighting factor ρ=0 between the buffer change and the complexity requirement, the initial bit budget of the sequence is Br, the initial buffer fullness is β ₁ , and the cycle starts from the first image group;

(3) Calculating the gth image group allocation bit number Bt(g);

(4) Use the tentatively allocated bit budget Bt(g) to detect the buffer state;

If the filling degree of the buffer zone is less than 80%, go to step (5), otherwise adjust ρ to increase it by 0.1, and repeat step (3);

(5) Update the buffer fullness B _g , and update the sequence remaining bit budget Br;

(6) Until the last GOP, the base layer GOP-level rate control ends.

Referring to Figure 4, the above-mentioned enhanced layer code rate control extracts the required bit rate code stream from the total code stream, and the specific steps are as follows:

(1) Set the required target bit rate Rt;

(2) Obtain the base layer bit rate R ₀ from the base layer code rate control process; when R ₀ > Rt, extracting the target bit stream fails and exits; otherwise proceed to the next step (3);

(3) The target bit rate is revised to Rt=Rt-R ₀ ;

(4) Calculate the bit rate R _F of a successive refinement layer;

(5) If R _F ≤ Rt, then modify the target bit rate to Rt=Rt-R _F , repeat step (4);

Otherwise, the refinement packet is truncated, and Rt=0, the target bit stream is successfully extracted, and the rate control of the enhancement layer ends.

Next, compare the algorithm in this paper with the code rate control algorithm of JM8.6+FGS. The simulation conditions are shown in Table 1.

Table 1. Encoding parameters Symbol Mode CABAC RD Optimization On GOP Structure IPPP... Reference Frame 1 MV Search Range 32

An example is given below. With the Foreman sequence, after the FGS encoding process, if the decoder wants to extract the CIF format, the target bit rate is 395kbit/s (352×288@30fps: 395kbit/s), then R _t =395kbit/s, R ₀ =158.5152 kbit/s. R ₀ <R _t , so the target code stream can be extracted from the total code stream. and the target bitrate is modified to $R_t\&DoubleLeftArrow;R_t-R_0=236.4848\mathrm{kbit}/\mathrm{the\; s}.$ Then the bit rate of the first successive refinement layer is calculated as R _F =68.1984 kbit/s. R _F < R _t , R _t = R _t − R _F =168.2864 kbit/s, continue to calculate the bit rate of the next successive refinement layer R _F =243.9704 kbit/s. R _F > R _t , so the successive refinement layer is truncated. R _t =0. The target bitrate bitstream is extracted successfully.

Table 2 shows the luminance PSNR of this method relative to JM8.6+FGS and the comparison with the target bit rate matching performance in the CIF format under the three test sequences of Foreman, Coastguard and News. Compared with JM8.6+FGS, the method of the present invention has an average Y-PSNR gain of 2.45dB under the CIF format. And more closely matches the target bitrate.

Table 2.3 sequence CIF format, comparison of two methods (352×288@30fps, 300Frames) Jm8.6+FGS Combined FGS of the present invention gain test sequence Target bit rate (kbit/s) Luminance peak signal to noise ratio (dB) Standard Deviation(dB) Actual bit rate (kbit/s) Luminance peak signal to noise ratio (dB) Standard Deviation(dB) Actual bit rate (kbit/s) Brightness Peak SNR Gain(dB) Standard deviation gain (dB) Foreman 395 32.11 1.53 394.3240 34.31 1.12 394.9824 2.20 -0.41 Coastguard 537 29.64 0.96 536.3872 31.94 1.01 536.9848 2.29 0.05 News 474 37.04 0.96 473.5320 39.89 0.79 473.9824 2.85 -0.17 average 2.45 -0.18

Claims (4)

1. bit rate control method based on H.264/AVC subtle granule telescopic coding, it is characterized in that encoder-side combines base layer stream and two code streams of enhancement layer stream, the bit rate control method that employing is encoded based on fine granularity scalable H.264/AVC, generate an ASCII stream file ASCII at receiving terminal, only needing to realize an ASCII stream file ASCII of transmission and very small amount of enhancement layer coding information; The steps include:

(1) encode based on subtle granule telescopic H.264/AVC: the result according to rate distortion comes self adaptation selected amount parameter, total bit stream of encoding;

(2) base layer stream rate control: the Rate Control of basic layer being done the image sets level in encoder-side;

(3) enhancement layer Rate Control: and enhancement layer is done the Rate Control that becomes more meticulous one by one in encoder-side;

(4) Rate Control of when decoding end extracts target bits stream from total bit stream, being done, the code stream of extraction arbitrary bit rate.

2. bit rate control method of encoding according to claim 1 based on subtle granule telescopic H.264/AVC, it is characterized in that described subtle granule telescopic coding based on H.264/AVC, all to carry out the motion-compensated time decomposable process to each space layer, thereby the time scalability is provided; Its concrete steps are as follows:

(1) input original series, promptly full precision temporal resolution sequence;

(2) time is decomposed predictive filtering, obtains high communication number, calculates its quantization parameter QPH; The high communication number of all these layers constitutes a time enhancement layer; Time is decomposed renewal filtering, obtains low-pass signal, calculates its quantization parameter QPL;

(3) with each reduction by 1/2 precision temporal resolution, repeat from top extremely last one deck (lowermost layer) (2) set by step, and all low-pass signals that obtain last one deck constitute basic layer of times;

(4) the residual error texture information is through residual coding, and the motion descriptor obtains the scalable quality layers that becomes more meticulous one by one of signal to noise ratio through motion encoded.

3. bit rate control method of encoding based on subtle granule telescopic H.264/AVC according to claim 1 is characterized in that the concrete steps of described basic layer bit rate control are as follows:

(1) estimates by calculating the complexity that the mean space complexity defines image sets;

(2) weighted factor ρ=0 between variation of initial setting buffering area and the complexity demand, the initial bit budget of sequence is Br, initial buffer district fullness degree is β ₁, begin circulation from first image sets;

(3) calculate g image sets allocation bit and count Bt (g);

(4) the bit budget Bt (g) with exploratory distribution detects buffer state;

If buffering area fullness degree＜80% then changes step (5) down, make it to increase progressively 0.1, repeating step (3) otherwise adjust ρ;

(5) upgrade buffering area fullness degree B _g, upgrade sequence remaining bits budget Br;

(6) until last image sets, then basic tomographic image group level bit-rate control finishes.

4. bit rate control method of encoding based on subtle granule telescopic H.264/AVC according to claim 1 is characterized in that described enhancement layer Rate Control, extracts required bit rate code stream from total code stream, and concrete steps are as follows:

(1) sets required target bit rate Rt;

(2) obtain basic layer bit rate R from basic layer bit rate control procedure ₀

Work as R ₀During＞Rt, extract the target bits loss and lose, withdraw from;

Otherwise carry out next step (3);

(3) target bit rate is revised as Rt=Rt-R ₀

(4) calculate the bit rate R of the layer that becomes more meticulous one by one _F

(5) if R _F≤ Rt, then the modifying target bit rate is Rt=Rt-R _F, repeating step (4);

Otherwise block the bag that becomes more meticulous, and Rt=0, to extract target bits and flow successfully, the enhancement layer Rate Control finishes.

Images