CN101677398A - Scene switching code rate control method - Google Patents

Abstract

The invention relates to a scene switching code rate control method which comprises the following steps: judging whether the current coded image occurs the scene switching or not; and calculating quantization parameters which are applicable to a scene switching frame and using the quantization parameters for encoding if the scene switching occurs. The method can be conveniently integrated in a variety of prior video encoding standards. Compared with a variety of prior methods, the differences are as follows: the method does not change a GOP structure for encoding, does not re-distribute code rate, and directly calculates the quantization parameters for the scene switching frame.

Description

Scene switching code rate control method

Technical field

The present invention relates to coding techniques, particularly scene switching code rate control method.

Background technology

For conserve transmission bandwidth and memory space, support application such as digital television broadcasting, remote monitoring, ordering digital video by short message, wireless multimedia communication, video coding has become one of focus of domestic and international research and commercial Application.So far formulated a series of video encoding standard in the world, as MPEG-1/2/4, H.261/H.263, H.264/AVC, and the domestic AVS video encoding standard of formulating at present in improving.Wherein, H.264/AVC compared with former standard with AVS, though coding framework is similar, its compression efficiency has all improved more than at least one times than any standard in the past, is adapted at more scene and uses.Fig. 1 is the coding framework of video coding.

The main encoding and decoding technique that present video encoding standard adopts comprises: integer transform, Rate Control, quantification, infra-frame prediction, inter prediction, variable-length encoding and block-eliminating effect filtering.

Wherein, the purpose of Rate Control is can reach the optimal image quality in order to guarantee producing under the condition of specifying the code check code stream, and guarantees that the encoding and decoding buffering area can not overflow.The Rate Control module mainly be output as quantization parameter (QP), in the cataloged procedure,, search (perhaps calculating) and obtain quantization step (QPStep) according to this QP value, be used for image quantization.

The key step of Rate Control comprises: at first, determine initial Q P, this can be that the user imports, and also can be that rate control algorithm calculates.But, use for great majority, need rate control algorithm to calculate automatically.Second step distributed for the position, promptly according to target bit rate and buffer state, for each two field picture distributes certain bit number; At last, calculate and adjust quantization parameter, the approaching bit number of distributing to this frame of bit number that feasible this two field picture of coding is used, thus reach target bit rate.Fig. 2 is the block diagram of Rate Control module.

Rate Control is a requisite technology in the video coding, any standard is left Rate Control, and its application all can be restricted, in the transmission system under constant bandwidth constraints, if there is not suitable bit rate control method, the transmission compressing video frequency flow that channel can't be stable.For this reason, every kind of video encoding standard has all been recommended a kind of Rate Control scheme that is fit to himself.

In the present rate control algorithm, use RQ forecast model and linear MAD forecast model, calculate quantization step.Need estimate to obtain the complexity PMAD of image according to the similitude of consecutive frame, and then calculate the quantization step that is used to encode according to the bit number that distributes.

Existing bit rate control method is based on the similar supposition of adjacent image content frame, does not consider the scene switch instances.

When scene was switched, because no longer there is similitude in consecutive frame, the RQ forecast model and the MAD forecast model that therefore are used for Rate Control lost efficacy, and its quantization parameter that calculates will be not suitable for the scene switch frame.

In addition, because no longer there is similitude in consecutive frame, when coding P frame and B frame, estimation will lose efficacy, and cause many macro blocks to adopt the frame mode coding, and code stream length and picture quality fluctuation are bigger.

Fig. 3 has provided the coding result of 512Kbps mixed C IF code stream under the AVS reference software bit rate control method, can see, when occurrence scene switches, because inappropriate quantization parameter causes picture quality to decline to a great extent.

Fig. 4 has provided the code length of 512Kbps mixed C IF code stream under the AVS reference software bit rate control method, can see in scene switching place, and the code length fluctuation is very big.

At present, scene switching code rate control method can be divided into 2 classes.The 1st class is 2 times coded systems, finds the scene switch frame in the first pass cataloged procedure, and record coding information, is used for coding second time.This class methods amount of calculation is big, and is not suitable for real-time application.

The 2nd class methods are 1 time coded system.Whether detect at first that occurrence scene switches, switch then carry out different Rate Control if judge on-the-spot scape.A kind of method is set at the I frame to scene switch frame type, just restarts a GOP coding, and this method has changed the structure of GOP, has increased the complexity of encoder to a certain extent.Additive method does not change gop structure, but is scene switch frame re-distribute code rate, calculates QP according to the code check that distributes then.

On the method for scene detection, 2 consecutive frame pixel value differences, image pixel mean value, and the estimation residual error is the scene detection parameter of using always.

Summary of the invention

Bit rate control method when the purpose of this invention is to provide a kind of scene and switching.

For achieving the above object, a kind of scene switching code rate control method comprises step:

A) judge whether occurrence scene switches current encoded image;

B), then calculate the quantization parameter that is fit to this scene switch frame, and use described quantization parameter to encode if occurrence scene switches.

Method of the present invention can be integrated in the existing various video encoding standard easily.Different with existing the whole bag of tricks, the present invention does not change the gop structure of coding, also re-distribute code rate not, but directly calculate the quantization parameter that is used for the scene switch frame.

Description of drawings

Fig. 1 is the encoder functionality block diagram;

Fig. 2 is the Rate Control block diagram;

Fig. 3 is a picture quality before and after scene is switched;

Fig. 4 is a code length;

Fig. 5 is the invention flow chart;

Fig. 6 is the scene detection flow process;

Fig. 7 is that scene switch frame quantization parameter calculates;

Fig. 8 is the RD performance;

Fig. 9 is that subjective quality compares;

Figure 10 is that code stream length compares.

Embodiment

Flow chart of the present invention, as shown in Figure 5:

At first, carry out 501 steps, the computed image feature, and judge whether that occurrence scene switches.

The concrete implementation of 501 steps is as shown in Figure 6:

In 601 steps, calculate the graded value of current encoded image horizontal direction:

MAGh wherein _cFor the horizontal direction gradient absolute value of current encoded frame and, MAGh _pFor the horizontal direction gradient absolute value of a last coded frame and, MAGh _cAnd MAGh _pComputational methods are identical, for:

$\mathrm{MAGh}=\underset{i}{\overset{W}{\mathrm{\Σ}}}\underset{j}{\overset{H}{\mathrm{\Σ}}}|I_{\mathrm{ij}}-I_{i,j+1}|,$ I _IjBe pixel value, W and H are the wide and high of image.

In 602 steps, calculate the graded value of current encoded image vertical direction:

MAGv wherein _cFor the vertical gradient absolute value of current encoded frame and, MAGv _pFor the vertical gradient absolute value of a last coded frame and, computational methods are

I _IjBe pixel value, W and H are the wide and high of image.

In 603 steps, compute gradient changes C=C _h+ C _v, and judge itself and preset threshold C _T1And C _T2Relation, if C＞C _T1, then the scene switching has taken place in present frame; If C＜C _T2, then present frame does not have occurrence scene to switch; If graded is between 2 setting thresholds, then execution in step 604, do further judgement.Here, C _T1And C _T2All greater than 0, C _T1Typical case's span between 0 to 10, the numerical value beyond also desirable this scope.C _T2Span between 0 to 1, C _T2Value needs less than C _T1

In the step 604, calculate current encoded image bit plane Change in Mean:

M wherein _cBe the bit plane average of current encoded frame, M _pBe the bit plane average of a last coded frame, computational methods are:

I _{Ij, k}Be k bit plane value of pixel, W and H are the wide and high of image.Here,, select different bit planes, for example can select the the 7th, the 6, the 5th bit plane according to different application; Also can select the the 5th, the 4, the 2nd bit plane.

In the step 605, judge bit plane Change in Mean and preset threshold C _T3Relation, if C _b＞C _T3, then the scene switching has taken place in present frame; Otherwise, think that then present frame does not have occurrence scene to switch.Here, C _T3For greater than any number of 0.

If in 501 steps, judge the scene switching taken place that then execution in step 502, for the scene switch frame calculates quantization parameter.

The concrete implementation of 502 steps is as shown in Figure 7:

In 701 steps, at first calculate scene switch frame characteristics of image

N _PixelBe number of pixels in the image.

In 702 steps,, calculate the quantization step that is used for the scene switch frame according to the characteristics of image that obtains

Here, α, beta, gamma are respectively parameter 1, parameter 2 and parameter 3, and these 3 parameter value scopes can be got arbitrary value without limits; Bpf calculates according to target bit rate and frame per second size, has represented the bit number of every frame under the target bit rate:

Bitrate is the coding target bit rate, and f is a frame per second; QPstep is for calculating quantization step.

In 703 steps,, the quantization step that calculates above is mapped as quantization parameter according to video encoding standard.Different video encoding standards has been stipulated the relation of quantization step and quantization parameter: QP=QPStep2QP (QPStep), and QPStep2QP finishes the function of quantization step QPStep to quantization parameter QP mapping in the video encoding standard here.

In 704 steps, the quantization parameter that calculates is adjusted, if current GOP residue bit number, the bit number less than needing according to target bit rate coding residual image then adds 1 with quantization parameter; Otherwise quantization parameter is constant.

In 705 steps, the quantization parameter that calculates is composed quantization parameter to current GOP, and among the GOP quantization parameter and, putting the P frame counter of having encoded among the GOP simultaneously is 0.

Be example with the AVS standard below, specify concrete enforcement of the present invention.Other embodiments, though do not comprise in the present embodiment, but still in right coverage of the present invention.

At first, calculate the graded value of current encoded image horizontal direction $C_h=\frac{|{\mathrm{MAGh}}_c-{\mathrm{MAGh}}_p|}{{\mathrm{MAGh}}_c};$

Second step, the graded value of calculating current encoded image vertical direction

$C_v=\frac{|{\mathrm{MAGv}}_c-{\mathrm{MAGv}}_p|}{{\mathrm{MAGv}}_c};$

In the 3rd step, compute gradient changes C=C _h+ C _v, and judge itself and preset threshold C _T1And C _T2Relation, if C＞C _T1, then the scene switching has taken place in present frame, carries out for the 6th step; If C＜C _T2, then present frame does not have occurrence scene to switch, according to original rate control algorithm coding present image; If graded between 2 setting thresholds, then carried out for the 4th step, do further judgement.In the present embodiment, C _T1And C _T2Difference value 0.8 and 0.2;

In the 4th step, calculate current encoded image bit plane Change in Mean Get the the 4th, the 5, the 6th bit plane in the present embodiment.

In the 5th step, judge bit plane Change in Mean and preset threshold C _T3Relation, if C _h＞C _T3, then the scene switching has taken place in present frame, continues to carry out the 6th step; Otherwise, think that then present frame does not have occurrence scene to switch, according to original rate control algorithm coding present image.In the present embodiment, C _T3Value 0.1;

In the 6th step, at first calculate scene switch frame characteristics of image

Calculate the quantization step that is used for the scene switch frame then

In the present embodiment, get parameter 1, parameter 2 and parameter 3 are respectively α=55000, β=20000, γ=1000;

In the 7th step,, the quantization step that calculates above is mapped as quantization parameter according to coding standard.In AVS, quantization step and quantization parameter are exponential relationship: QP=Int (11.5 * ln (QPStep)); In H.264, the pass of quantization step and quantization parameter is:

At last, judge current GOP residue bit number, whether less than the bit number that needs according to target bit rate coding residual image.If less than, then the quantization parameter that the 7th step was calculated adds one; Otherwise, keep quantizing parameter constant.And the quantization parameter that calculates composed quantization parameter to current GOP, and among the GOP quantization parameter and, putting the P frame counter of having encoded among the GOP simultaneously is 0.

The bit rate control method that method of the present invention and AVS reference software realize contrasts, utilize 9 CIF size 30Hz standard sequence Foreman, paris, bus, football, news, mobile, tempete, mother_daughter and deadline mix 360 frame sequences that comprise 17 scene switchings.In these 9 sequences, football comprises strenuous exercise; Object exists translation and rotation among mobile and the bus, and also there is relative motion in background; Paris and deadline are complex background, the local motion sequence; Mother_daughter is simple background, the simple motion sequence; There is background motion in news; Foreman exists camera lens to rock; There is irregular rapid movement in tempete.These sequences are the cycle tests that AVS working group is recommended, and our these 9 sequences of arrangement at random splice 360 frame mixed sequences.

Utilize AVS reference software and this patent method, respectively at 384Kbps, 512Kbps compresses under 10000Kbps and the 1500Kbps target bit rate, and the results are as follows:

The RD curve can see that the method for this patent on average improves PSNR about 1 dB as shown in Figure 8.

On subjective quality, the present invention has well solved the problem that picture quality declined to a great extent when scene was switched, and is example with 512Kbps mixing compressed bit stream equally, and Fig. 9 left side is the compression effectiveness of AVS reference software, and the right is a compression effectiveness of the present invention.

On code stream length, the code stream length sudden change when the present invention has avoided scene to switch, Figure 10 has provided AVS reference software and compressed bit stream length of the present invention comparison under the 512Kbps target bit rate.

Claims (16)

1. scene switching code rate control method comprises step:

A) judge whether occurrence scene switches current encoded image;

B), then calculate the quantization parameter that is fit to this scene switch frame, and use described quantization parameter to encode if occurrence scene switches.

2. method according to claim 1 is characterized in that described step a) comprises:

Calculate current encoded frame and previous coded frame horizontal direction graded;

Calculating current encoded frame and previous coded frame vertical gradient changes;

If horizontal direction and vertical gradient change sum greater than threshold value C _T1, then occurrence scene switches.

3. method according to claim 2 is characterized in that also comprising:

If horizontal direction and vertical gradient change sum less than threshold value C _T1Greater than threshold value C _T2, then calculate current encoded frame and previous coded frame bit plane Change in Mean;

If the bit plane Change in Mean is greater than threshold value C _T3, then occurrence scene switches.

4. method according to claim 2 is characterized in that being calculated as follows the graded value of current encoded image horizontal direction:

$C_h=\frac{|{\mathrm{MAGh}}_c-{\mathrm{MAGh}}_p|}{{\mathrm{MAGh}}_c},$ MAGh wherein _cFor the horizontal direction gradient absolute value of current encoded frame and, MAGh _pFor the horizontal direction gradient absolute value of a last coded frame and.

5. method according to claim 2 is characterized in that being calculated as follows the graded value of current encoded image vertical direction:

$C_v=\frac{|{\mathrm{MAGv}}_c-{\mathrm{MAGv}}_p|}{{\mathrm{MAGv}}_c},$ MAGv wherein _cFor the vertical gradient absolute value of current encoded frame and, MAGv _pFor the vertical gradient absolute value of a last coded frame and.

6. method according to claim 3 is characterized in that being calculated as follows current encoded image bit plane Change in Mean value:

$C_b=\frac{|M_c-M_p|}{M_c},$ M wherein _cBe the bit plane average of current encoded frame, M _pBit plane average for a last coded frame.

7. method according to claim 2 is characterized in that described threshold value C _T2Value is between 0-1.

8. method according to claim 3 is characterized in that described threshold value C _T2Less than C _T1

9. method according to claim 3 is characterized in that described threshold value C _T3Greater than zero.

10. method according to claim 3 is characterized in that described threshold value C _T1Be 0.8.

11. method according to claim 8 is characterized in that described threshold value C _T2Be 0.2.

12. method according to claim 9 is characterized in that described threshold value C _T3Be 0.1.

13. method according to claim 1 is characterized in that described step b) comprises:

Calculate the present encoding frame image features;

Calculate the quantization step of current encoded frame;

Calculate the quantization parameter of current encoded frame;

Adjust quantization parameter according to residue bit number;

Upgrade the GOP quantization parameter.

14. method according to claim 13 is characterized in that being calculated as follows the present encoding frame image features:

${\mathrm{MAG}}_c=\frac{{\mathrm{MAGh}}_c+{\mathrm{MAGv}}_c}{{2N}_{\mathrm{pixels}}},$ N _PixelBe number of pixels in the image.

15. method according to claim 13 is characterized in that being calculated as follows the quantization step of current encoded frame $\mathrm{QPstep}=\frac{\mathrm{\α}\×{\mathrm{MAG}}_c+\mathrm{\β}}{\mathrm{bpf}+\mathrm{\γ}},$ Here, α, beta, gamma are respectively parameter 1, parameter 2 and parameter 3; Bpf calculates according to target bit rate and frame per second size, has represented the bit number of every frame under the target bit rate: $\mathrm{bpf}=\frac{\mathrm{bitrate}}{f},$ Bitrate is the coding target bit rate, and f is a frame per second; QPstep is for calculating quantization step.

16. method according to claim 15 is characterized in that described parameter alpha=55000, β=20000, γ=1000.

Images