CN1870753A - Integration code rate control method of low complexity - Google Patents

Abstract

This invention relates to a method for controlling the integral code rate of low complexity, which first of all, distributes ralated bit number of each frame in the GOP, then distributes bit numbers for the current frames in the GOP then evaluates the quantized parameters of a macroblock needed by coding based on the integral code rate control technology, then feeds back the quantized parameters to the rate distortion optimizing technology and other code parts, in which, many models are used including a complexity evaluation model, a liquid flow barrage model and a integral code rate control model.

Description

A kind of integration code rate control method of low complex degree

Technical field

The present invention relates to a kind of integration code rate control method of low complex degree, particularly a kind of bit rate control method that is applicable to the low complex degree of newest standards encoder belongs to compression of digital video, digital video communication technique field.

Background technology

Video compression is a kind of lossy compression method, in the spatial redundancy and time redundancy in eliminating video sequence, by getting rid of the part high fdrequency component in the video information, reaches very high compression ratio.Wherein, the main means of adjusting compression ratio are by adjusting the size of quantization step, residual image is quantized, producing different bit numbers.The Rate Control part output by selecting suitable quantization step to come the control bit number just in the video encoder for encoder.As seen, Rate Control is the video encoder important component part.In cataloged procedure, Rate Control needs the requirement according to the network bandwidth, when control output bit number changes, guarantees the quality of output video as far as possible.

Till now, many bit rate control methods have appearred, the wherein more famous TM5 that MPEG-2 is arranged (Test Model 5) bit rate control method, TMN H.263 8 (Test Model Near 8) bit rate control method, the VM8 of MPEG-4 (Verification Model version 8) bit rate control method etc.

H.264/MPEG-4AVC up-to-date in the world video compression standard is now.The standard of latest domestic is the AVS video compression standard of formulating.These two standards have adopted up-to-date in recent years technology.Both are different aspect some, but most of technology is very close.Wherein, the both has adopted an important techniques---rate-distortion optimization technology.This technology has obtained extraordinary trading off between bit rate output and video quality distortion rate.But this technology and traditional bit rate control method be contradiction to some extent again.This contradiction is: traditional bit rate control method need be known the residual error of current macro before the required quantization parameter of calculation code current macro; And the rate-distortion optimization technology need be known the required quantization parameter of coding current macro before calculating the residual error of current macro.H.264 the bit rate control method in the reference software, successful solution this contradiction.But cost is to have brought very high complexity.

Summary of the invention

Main purpose of the present invention provides a kind of simply and effectively bit rate control method---integration code rate control method.This bit rate control method is not only applicable to traditional MPEG-2, MPEG-4, encoder H.263, and be equally applicable to up-to-date H.264, the AVS encoder.For the encoder that meets newest standards.This method can effectively be avoided the contradiction between above-mentioned and the rate-distortion optimization technology, control more accurately bit rate output variation, obtain the video output of better quality; Simultaneously, its implementation complexity is very low, greatly reduces the demand to resource.

The objective of the invention is such realization:

A kind of integration code rate control method of low complex degree comprises the steps: at least

Step 1: for current group of picture (Group of Picture is called for short GOP) is distributed certain bit number;

Step 2: for P, the B frame of GOP distributes corresponding bit number, for the I frame of GOP is specified a quantization parameter;

Step 3: at each macro block, adopt the method for integration code rate control, calculate the quantization parameter of Rate Control.

Being implemented as follows of above-mentioned steps 1: before the j frame of i GOP of coding, remaining unappropriated bit number among i the GOP:

$B_i\left(j\right)=\left\{\begin{array}{cc}\frac{R_i\left(j\right)}{f}\×N_f-V_i\left(j\right),& j=1,\\ B_i(j-1)+\frac{R_i\left(j\right)-R_i(j-1)}{f}\×N_i-j+1)-b_i(j-1),& j=\mathrm{2,3},...,N_i.\end{array}\right.---\left(1\right)$

Here B _i(j) remain bit number to be allocated when beginning to encode i GOP j frame; R _i(j) be instantaneous code check when coding j frame; N _iIt is the frame number among i the GOP; F is the coding frame per second; V _i(j) be the full scale (actual buffer fullness) of virtual buffering region; b _i(j-1) be the actual bit number that the j-1 frame is produced.

In fact, under the constant code rate situation of (Constant Bit Rate is called for short CBR), R _i(j) equal R _i(j-1).This up-to-date style (1) can be reduced to

$B_i\left(j\right)=\left\{\begin{array}{cc}\frac{R_i\left(j\right)}{f}\×N_i-V_i\left(j\right),& j=1,\\ B_i(j-1)-b_i(j-1),& j=\mathrm{2,3},...,N_i.\end{array}\right.---\left(2\right)$

V in the formula (1) _i(j) more new formula is

$V_i\left(j\right)=\left\{\begin{array}{cc}0,& i=1,j=1,\\ V_{i-1}\left(N_{i-1}\right),& i\≠1,j=1,\\ V_i(j-1)+b_i(j-1)-\frac{R_i(j-1)}{f},& j=\mathrm{2,3},...,N_i.\end{array}\right.---\left(3\right)$

I frame when beginning for each GOP is for it specifies a quantization parameter.

(1) to first frame (I frame) of first GOP, need situation in conjunction with image size and bandwidth, specify a rational quantization parameter QP1 (1) from the outside for it.Under the default situations (corresponding encoder H.264)

$Q{P}_1\left(1\right)=\left\{\begin{array}{cc}40,& \mathrm{bpp}\&le;l1\\ 30,& l1<\mathrm{bpp}\&le;l2,\\ 20,& l2<\mathrm{bpp}\&le;l3,\\ 10,& \mathrm{bpp}>l3.\end{array}\right.---\left(4\right)$

Here bpp (bpp:bits per pixel) is the used bit number of each pixel, $\mathrm{bpp}=\frac{R_1\left(1\right)}{f\&times;N_{\mathrm{pixel}}},$ N _PixelBe the counting of pixel in the two field picture.When the QCIF/CIF situation, specify l1=0.15 here, l2=0.45, l3=0.9 specifies l1=0.6, l2=1.4, l3=2.4 when greater than top picture size.

(2) for first frame (I frame) of i GOP, the value of its quantization parameter is as follows

$Q{P}_i\left(1\right)=\max \{Q{P}_{i-1}\left(1\right)-2,\min \{Q{P}_{i-1}\left(1\right)+2,\frac{\mathrm{SumQP}(i-1)}{N_p(i-1)}-1\left\}\right\}---\left(5\right)$

Determine, wherein:

N _p(i-1): the frame number of the P frame of i-1GOP;

SumQP (i-1): the average quantisation parameter sum of the P frame of i-1GOP,

For the quantization parameter of the I frame of first GOP by the user according to bandwidth, picture size is set an initial value.

Following formula is the result further be adjusted to

QP _i(1)＝QP _i(1)-1，if QP _i(1)＞QP _i-1(N _i-1-L)-2 (6)

QP _I-1(N _I-1-L) be the quantization parameter of last P frame among the last GOP; L is the frame number of two P interframe B frames

Described step 2 specific implementation is: be not all each frame allocation bit number according to frame type:

(1) calculates in current GOP virtual buffering region target full scale (target buffer level) Si (j) corresponding to each P frame or B frame

Behind first I frame of the current GOP that encoded, before first P frame of encoding, virtual buffering region initial target full scale is

S _i(2)＝V _i(2) (7)

For follow-up frame, virtual buffering region target full scale computing formula is

$S_i(j+1)=S_i\left(j\right)-\frac{S_i\left(2\right)}{N_i-2}---\left(8\right)$

Comprehensive above two formulas, virtual buffering region target full scale is calculated by following formula:

$S_i\left(j\right)=\left\{\begin{array}{cc}{V}_i\left(2\right),& j=2,\\ S_i(j-1)-\frac{S_i\left(2\right)}{N\left(i\right)-2},& j>2.\end{array}\right.---\left(9\right)$

(2) be current P frame or B frame allocation bit number

At first,, calculate the target bit of j frame according to virtual buffering region target full scale, virtual buffering region full scale, frame per second, current channel width, as shown in the formula

${\tilde{T}}_i\left(j\right)=\frac{R_i\left(j\right)}{f}+\mathrm{\&gamma;}\&times;(S_i\left(j\right)-V_i\left(j\right))---\left(10\right)$

Here γ is a constant, is 0.5.

According to the remaining target bit of current GOP, be present frame allocation bit number again

$T_{t,i}\left(j\right)=\frac{W_{t,i}(j-1)\&times;B_i\left(j\right)}{W_{p,i}(j-1)\&times;N_{p,r}+W_{b,i}(j-1)\&times;N_{b,r}11}---$

Wherein: t represents P frame or B frame; N _{P, r}And N _{B, r}Be respectively the frame number that remains P frame to be encoded and B frame among the current GOP.Here W _{P, i}(j) be the average complexity of P frame, W _{B, i}(j) be the average complexity of B frame.Their computing formula is

$W_{p,i}\left(j\right)=\frac{b_i\left(j\right)\&times;Q{P}_{p,i}\left(j\right)}{K_p}$

$W_{b,i}\left(j\right)=\frac{b_i\left(j\right)\&times;{\mathrm{QP}}_{b,i}\left(j\right)}{K_b}$

Here K _p, K _bBe the constant relevant with quantification mechanism, when encoder is the MPEG-2 encoder, K _p=1.0, K _b=1.4; When encoder is H.264 encoder, K _p=1.0, K _b=1.3636.

Take all factors into consideration formula (10), (11), obtain the target bit rate of current P frame or B frame

$T_i\left(j\right)=\mathrm{\&beta;}\&times;{\hat{T}}_i\left(j\right)+(1-\mathrm{\&beta;})\&times;{\tilde{T}}_i\left(j\right)---\left(12\right)$

Wherein β is a constant, is 0.5.

At last, consider HRD (corresponding to H.263, H.264, being VBV, is BBV) in MPEG series in AVS, target bit is limited within the specific limits.

The frame of having encoded, the actual bit that is produced is added in the buffering area, for the buffering area degree of taking after guaranteeing to upgrade is unlikely to too high, adopts certain frame-skipping strategy.

Described step 3 specific implementation is: use integral control method and be in the frame each macroblock allocation code check and calculate the corresponding quantization parameter

For the I frame, the quantization parameter of all macro blocks all equates in the frame.Therefore macro-block level bit rate control method described here is at P frame and B frame.

The macro-block level bit rate control method is that each macro block in the frame is selected suitable quantization parameter in the encoder encodes process, and the bit number that makes whole frame produce like this is close with the code check bit number that distributes for present frame.

Be the detailed description of macro-block level integral control method below

(1) adopt integral control method to calculate the quantization step of current macro

Qstep _t+＝error_bits _t/kiconst (13)

The t here represents P frame or B frame.Error_bits _tBe the true bit number of the last same type of mb gained of coding and the difference of target bit; Kiconst is an integral constant; Qstep _tBe the required quantization step of coding current macro, its initial value is 10.According to the mapping ruler of encoder quantization step Qstep _tChange into corresponding quantization parameter QP _{L, i}(j).For not influencing video quality, the variation of macroblock quantization parameter should be slow continuously in the same frame, to QP _{L, i}(j) excursion is made following restriction QP _{L, i}(j)=max{QP _{L-1, i}(j)-and DQuant, min{QP _{L, i}(j), QP _{L-1, i}(j)+DQuant}} (14)

Here, DQuant is the quantization parameter increment.Because the quantification mechanism of different coding system is different, thereby the value of DQuant generally also is different, gets 1 here.

In addition, the bit number that has produced at present frame is during more than or equal to the target bit of present frame, and the quantization parameter of current macro is pressed the following formula value:

QP _l，i(j)＝QP _l-1，i(j)+DQuant (15)

At last, for guaranteeing level and smooth on the video pictures visual quality, to the QP that tries to achieve above _{L, i}(j) do following qualification:

QP _l，i(j)＝max{QPMIN， QP _t，i(j)-DQP，min{QP _MAX， QP _t，i(j)+DQP，QP _l，i(j)}}(16)

Here QP _MINBe minimum quantization parameter, QP _MAXBe maximum quantization parameter; QP _{T, i}(j) be the average quantisation parameter value of same type previous frame; DQP for the current macro quantization parameter can with the maximum difference of the average quantisation parameter value of previous frame of the same type;

(2) to current macro encode (H.264, in the encoder of AVS, carry out rate-distortion optimization and coding to current macro this moment)

(3) upgrade present frame residue target bit, upgrade the bit difference of current macro

error_bits _t＝b _l，i(j)-T _i(j)/mb_cn (17)

b _{L, i}(j) be the coding bit number that current block produced; T _i(j)/and mb_cn: the target bit of current macro; T _i(j) be the target bit rate of present frame; Mb_cn is the macroblock number that current video sequence one two field picture is comprised.

Description of drawings:

Fig. 1 is the video coding system block diagram;

Fig. 2 is the video coding system main program flow chart;

Fig. 3 is the key step flow chart of the inventive method implementation procedure.

Embodiment:

The objective of the invention is to: under the situation of the quality that guarantees output video, by the output that comes the control bit number for the suitable quantization step of digital video code selection, with adapt to the network bandwidth and requirement.In the video coding system block diagram shown in Fig. 1, video sequence represents that it is as the input of video coding system by the original video sequence that does not compress of other equipment acquisitions.At the video encoder shown in Fig. 1 can be video coding chip or video coding program, by it original video sequence is carried out compressed encoding, the stream file of generated code as a result of compression.The data volume of this ASCII stream file ASCII is much smaller than original video sequence, thereby reached the purpose of compressing video information.

Shown in Figure 2 is the main program flow chart of video coding system.Except that this part of Rate Control, the present invention has adopted international video encoding standard basic coding framework H.264JM86.

Shown in Figure 3 is the realization flow figure of Rate Control part.Bit rate control method is a core of the present invention, and it is not only applicable to traditional MPEG-2, MPEG-4, encoder H.263, and be equally applicable to up-to-date H.264, the AVS encoder.

In order to verify actual performance of the present invention and code efficiency, carried out following contrast experiment.To identical video sequence, adopt identical coding parameter, carry out Rate Control program of the present invention and Rate Control program H.264JM86 respectively.Table 1,2 is respectively the comparison of mean P SNR value under different target bit rate situations.As can be seen from the table, improve algorithm under low code check situation, better than former algorithm, under other situations, the mean P SNR value of its mean P SNR value and former algorithm remains basically stable.The actual bit rate and the target bit rate that improve algorithm output are very approaching, and can change with the variation of target bit rate.The present invention simultaneously proposes the level that bit rate control method can maintain the full scale of virtual buffering region an appropriateness, guarantees neither also underflow not of overflow of buffering area.

Each sequence average PSNR value before and after improving when table 1 hangs down code check

Video name	Form	Frame number	Code check (kbps)	PSNR(dB)
							Former algorithm	After the improvement	Increase
				news	QCIF	300				16	33.639	33.871	0.23
64	40.133	40.258	0.13
							foreman	QCIF	300	16	30.382	30.671	0.29
64	35.879	35.946	0.07
				bus	QCIF	150				16	24.097	24.468	0.37
64	28.563	28.479	-0.08
							pairs	QCIF	300	16	28.254	28.111	-0.13
64	33.783	33.786	0.00
				silent	QCIF	300				16	33.007	33.267	0.26
64	39.132	39.334	0.20
							tempet	QCIF	250	16	24.769	25.177	0.41
64	29.142	29.273	0.13

Each sequence average PSNR value before and after improving during code check in the table 2

Video name	Form	Frame number	Code check (kbps)	PSNR(dB)
							Former algorithm	After the improvement	Increase
				news	QCIF	300				256	38.817	39.124	0.31
512	42.654	42.708	0.05
							foreman	QCIF	300	256	33.888	34.070	0.18
512	36.922	36.949	0.03
				bus	QCIF	150				256	27.483	27.449	-0.03
512	30.795	30.545	-0.25
							pairs	QCIF	1065	256	33.393	33.878	0.48
512	37.438	37.658	0.22
				silent	QCIF	300				256	36.227	36.370	0.14
512	39.759	39.785	0.03
							tempet	QCIF	260	256	28.363	28.780	0.42
512	31.353	31.400	0.05

Claims (17)

1. the integration code rate control method of a low complex degree is characterized in that, comprises the steps: at least

Step 1: for current group of picture GOP distributes certain bit number;

Step 2: for P, the B frame of GOP distributes corresponding bit number, for the I frame of GOP is specified a quantization parameter;

Step 3: at each macro block, adopt the method for integration code rate control, calculate the quantization parameter of Rate Control.

2. integration code rate control method according to claim 1 is characterized in that, described step

1) target bit of current GOP has following formula to determine

$B_i\left(j\right)=\left\{\begin{array}{cc}\frac{R_i\left(j\right)}{f}\&times;N_i-V_i\left(j\right),& j=1,\\ B_i(j-1)+\frac{R_i\left(j\right)-R_i(j-1)}{f}\&times;N_i-j+1)-b_i(j-1)& j=\mathrm{2,3},...,N_i.\end{array}\right.$

Wherein: B _i(j): remain bit number to be allocated when beginning to encode i GOP j frame; R _i(j): the instantaneous code check when coding j frame; N _iIt is the frame number among i the GOP; F: coding frame per second; V _i(j): the full scale of virtual buffering region; b _i(j-1): the actual bit number that the j-1 frame is produced.

3. integration code rate control method according to claim 2 is characterized in that V _i(j) more new formula is $V_i\left(j\right)=\left\{\begin{array}{cc}0,& i=1,j=1,\\ V_{i-1}\left(N_{i-1}\right),& i\&NotEqual;1,j=1,\\ V_i(j-1)+b_i(j-1)-\frac{R_i(j-1)}{f},& j=\mathrm{2,3},...,N_i.\end{array}\right.$

4. integration code rate control method according to claim 1 is characterized in that, described step 2) in be that the corresponding bit number of appointment of P frame or B frame is by formula $T_i\left(j\right)=\mathrm{\&beta;}\&times;{\hat{T}}_i\left(j\right)+(1-\mathrm{\&beta;})\&times;{\hat{T}}_i\left(j\right)$ Determine that wherein: β is a constant, value is 0.5.

5. integration code rate control method according to claim 4 is characterized in that,

By formula ${\hat{T}}_{t,i}\left(j\right)=\frac{W_{t,i}(j-1)\&times;B_i\left(j\right)}{W_{p,i}(j-1)\&times;N_{p,r}+W_{b,i}(j-1)\&times;N_{b,r}}$ Determine, wherein:

T: represent P frame or B frame;

N _{P, r}: the frame number of residue P frame to be encoded among the current GOP;

N _{B, r}: the frame number of residue B frame to be encoded among the current GOP;

W _{P, i}(j): the average complexity of P frame;

W _{B, i}(j): the average complexity of B frame.

6. integration code rate control method according to claim 5 is characterized in that, P frame average

Complexity W _{P, i}(j) and

The average complexity W of B frame _{B, i}(j) respectively by formula $W_{p,i}\left(j\right)=\frac{b_i\left(j\right)\&times;Q{P}_{p,i}\left(j\right)}{K_p}$ With $W_{b,i}\left(j\right)=\frac{b_i\left(j\right)\&times;Q{P}_{b,i}\left(j\right)}{K_b},$ Wherein:

K _p, K _b: the constant relevant with quantification mechanism, when encoder is the MPEG-2 encoder,

K _p=1.0, K _b=1.4; When encoder is H.264 encoder, K _p=1.0,

K _b＝1.3636。

7. integration code rate control method according to claim 4 is characterized in that, By formula ${\tilde{T}}_i\left(j\right)=\frac{R_i\left(j\right)}{f}+\mathrm{\&gamma;}\&times;(S_i\left(j\right)-V_i\left(j\right))$ Determine that wherein: γ: be a constant, value is 0.5.

8. integration code rate control method according to claim 7 is characterized in that S _i(j) by formula $S_i\left(j\right)=\left\{\begin{array}{cc}{V}_i\left(2\right),& j=2,\\ S_i(j-1)-\frac{S_i\left(2\right)}{N\left(i\right)-2},& j>2.\end{array}\right.$ Determine.

9. integration code rate control method according to claim 1 is characterized in that, described step

The quantization parameter of the I frame of the GOP 2) is by formula

$Q{P}_i\left(1\right)=\max \{Q{P}_{i-1}\left(1\right)-2,\min \{Q{P}_{i-1}\left(1\right)+2,\frac{\mathrm{SumQP}(i-1)}{N_p(i-1)}-1\left\}\right\}$ Determine, wherein: N _p(i-1): the frame number of the P frame of i-1GOP; SumQP (i-1): the average quantisation parameter sum of the P frame of i-1GOP, for the quantization parameter of the I frame of first GOP by the user according to bandwidth, picture size is set an initial value.

10. integration code rate control method according to claim 1 is characterized in that, is initial value for the quantization parameter of the I frame of first GOP, and (corresponding encoder H.264) uses following formula to determine under the default situations

$Q{P}_1\left(1\right)=\left\{\begin{array}{cc}40,& \mathrm{bpp}\&le;l1,\\ 30,& l1<\mathrm{bpp}\&le;l2,\\ 20,& l2<\mathrm{bpp}\&le;l3,\\ 10,& \mathrm{bpp}>l3.\end{array}\right.$

Here $\mathrm{bpp}=\frac{R_1\left(1\right)}{f\&times;N_{\mathrm{pixel}}},$ N _PixelBe the counting of pixel in the two field picture, when the QCIF/CIF situation, specify l1=0.15 here, l2=0.45, l3=0.9 is when the size of coded image is specified l1=0.6, l2=1.4, l3=2.4 during greater than top picture size.

11. integration code rate control method according to claim 1 is characterized in that, the quantization parameter corresponding quantitative step-length in the described step 3) is by formula Qstep _t+=error_bits _t/ kiconst is definite, wherein:

T: represent P frame or B frame,

Error_bits _t: the bit number of a last actual generation of same type of mb and the difference of target bit;

Kiconst: integral constant;

Qstep _t: coding current macro required quantization step, its initial value is 10, according to the mapping ruler of encoder quantization step Qstep _tChange into corresponding quantization parameter QP _{L, i}(j).

12. integration code rate control method according to claim 10 is characterized in that, continuous macroblock quantization parameter QP in the frame _{L, i}(j) excursion is by formula QP _{L, i}(j)=max{QP _{L-1, i}(j)-and DQuant, min{QP _{L, i}(j), QP _{L-1, i}(j)+DQuant}} is definite, wherein:

DQuant: be the quantization parameter increment, the quantification mechanism of different coding system is different,

The value of DQuant generally also is different, gets 1 here.

13. integration code rate control method according to claim 1 is characterized in that, the bit number that the quantization parameter of each macro block in the described step 3) has produced at present frame is during more than or equal to the target bit of present frame, by QP _{L, i}(j)=QP _{L-1, i}(j)+DQuant is definite.

14., it is characterized in that final quantization parameter QP according to claim 10,11,12 described integration code rate control methods _{L, i}(j) by formula

QP _{L, i}(j)=max{QP _MIN, QP _{T, i}(j)-and DQP, min{QPMAX, QP _{T, i}(j)+and DQP, QP _{L, i}(j) } } determine, wherein:

QP _MIN: the minimum quantization parameter;

QP _MAX: maximum quantization parameter;

QP _{T, i}(j): the average quantisation parameter value of same type previous frame;

DQP: the current macro quantization parameter can with the maximum difference of the average quantisation parameter value of previous frame of the same type.

15. integration code rate control method according to claim 10 is characterized in that, after the macro block of having encoded, the difference of actual bit number that produces and target bit is by formula error_bits _t=b _{L, i}(j)-T _i(j)/mb_cn is definite, wherein:

b _{L, i}(j): the coding bit number that current macro produced;

T _i(j)/and mb_cn: the target bit of current macro;

T _i(j): the target bit rate of present frame;

Mb_cn: the macroblock number that current video sequence one two field picture is comprised.

Images