CN1992898A - Video rate control method with low complexity - Google Patents

Abstract

The invention relates to a simple video code rate control method, which comprises bit rate preset process and quantified parameter adjustment, wherein the bit rate preset process uses video sequence character to preset the code rate; the quantified parameter adjustment uses the video sequence character to predict the code complexity, to evaluate the quantified parameters; the video sequence character is the differential diagram representing the local motion of video sequence. The invention has the advantages that: it can adjust the code rate distribution via the change to video content; the quantified parameters have simple calculation, with adaptive content; the code distribution and adjustment can match the change of video content; and it uses accumulation refresh statistic method to avoid time delay.

Description

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

Technical field

The present invention relates to a kind of Rate Control mechanism, more particularly, the present invention relates to a kind of video code rate control method of the low complex degree based on the original video statistical nature.

Background technology

The Rate Control technology is hot issue of the video coding and the communications field, plays crucial effect in video communication.The Rate Control technology generally can be divided into three kinds: the one, utilize the analytic modell analytical model of code check one quantization parameter, and the 2nd, utilize code check is regulated in the control of encoding buffer, the 3rd, fully utilize above-mentioned two class methods.The advantage of first kind method is that code rate adjustment can be with separating that the method for resolving be optimized, but because the randomness of video content itself, single analytic modell analytical model can not accurately be portrayed the characteristic of video itself, so, need more Control Parameter and just can obtain satisfied effect, complexity is very high.Second class methods are simple to operate, but control precision is not enough.The 3rd class methods combine analytic modell analytical model and buffering area control, are most popular classes in the present Rate Control technology.

In recent years, along with the development of video coding technique, (H.264, VC1.0 AVS), correspondingly also has a series of rate control algorithms to propose to have produced new coding standard.But these rate control algorithms all are the 3rd class algorithm recited above usually, promptly based on the model of analytic modell analytical model and buffering area control.Though such model effectively combines analytic modell analytical model and buffering area controlling models, from rate control accuracy, can meet the demands, but because video sequence itself is difficult to by analytic modell analytical model accurately predicting (complexity height), so, existing method can not accurately control code check according to the intrinsic rate-distortion of video sequence (R-D) attribute itself, but video sequence regard as one stably random process control code check.But video sequence itself is the random process of a non-stationary, can make that like this R-D performance of existing bit rate control method is not high.

If overcome the deficiency of existing method, must promptly and accurately obtain the variation of video sequence, and come corresponding adjustment code check according to this variation.And the variation of video sequence can obtain by some statistics of video sequence itself.This tittle has generally comprised: image variance, consecutive frame histogram of difference, residual image variance and residual image histogram etc.Wherein consecutive frame histogram of difference (hereinafter to be referred as histogram of difference) can the interior power with respect to the reference frame movement degree of fine reflecting video frame.

Summary of the invention

The objective of the invention is to overcome the deficiencies in the prior art, provide a kind of rate-distortion performance high low complex degree code check control algolithm.

For achieving the above object, the video code rate control method of low complex degree provided by the invention comprises bit rate pre-allocation process and quantization parameter adjustment process; It is characterized in that described bit rate pre-allocation process is to utilize video sequence characteristics to carry out code check preassignment; Described quantization parameter adjustment process is to come the predictive coding complexity according to video sequence characteristics, and then carries out quantization parameter and estimate; Described video sequence characteristics is meant the histogram of difference of representing the video sequence local motion.

In the technique scheme, described code check preassignment comprises the steps:

Step 1): target setting code check (in the video coding specified nominal code check) and destination buffer occupancy (the specified occupancy of video encoder buffering area) and gop size; According to the histogram of difference of first each coding unit of inter-frame, the histogram of difference value of present frame is asked in initialization, and histogram mean value is initialized as this value; In same image sets, the mean value of successive image histogram of difference obtains according to the mode of cumulative mean;

Step 2): to each frame, the encoder complexity of first coding unit of initialization;

Step 3): calculate the difference of consecutive frame and add up its histogram, calculate the average of histogram of difference sequence; According to the difference between present frame histogram and the average, combining target code check and destination buffer occupancy are distributed frame level bit-rate according to average-scale model then;

Step 4): on the basis of frame one-level Data Rate Distribution,, adopt average-scale model to distribute the code check of basic cell level further according to the histogram of difference of each Rate Control elementary cell.

In the technique scheme, described quantization parameter is estimated to comprise the steps:

Step 5): according to adjacent encoder unit histogram distribution in the frame, the encoder complexity of prediction present encoding unit;

Step 6): obtain after the encoder complexity of present encoding unit,, calculate the quantization parameter of present encoding unit according to preallocated code check; Utilize this quantization parameter that present image is encoded, and go to revise the Data Rate Distribution of next frame with the code check that obtains behind the coding.

In the technique scheme, described Rate Control elementary cell is a macro-block line.

The invention has the advantages that:

1, can in time regulate Data Rate Distribution according to the variation of video content;

2, quantization parameter calculates simply effectively, has content-adaptive;

3, the video content variation is pressed close in Data Rate Distribution and adjusting, has good rate-distortion (R-D) performance;

4, adopt the mode of the Pleistocene series metering that adds up, avoided delay.

Description of drawings

Fig. 1 is the rate-allocation procedure schematic diagram;

Fig. 2 is the flow chart of quantization parameter estimating part;

Fig. 3 is the flow chart of bit rate control method overall process of the present invention.

Embodiment

The invention will be further described below in conjunction with accompanying drawing.

Rate-allocation procedure relates generally to histogram of difference and calculates, and frame level bit-rate distributes and the coding unit level bit-rate distributes.As shown in Figure 1, three parts refer to histogram calculation respectively among the figure, and frame distributes and the unit distributes.

Quantization parameter is estimated as shown in Figure 2.In code rate control method, in order to improve control precision, need to reduce the granularity of control, so every frame is divided into several basic coding unit, the Rate Control process is carried out on the basis of elementary cell.To each elementary cell, come the predictive coding complexity by its histogram of difference, just can obtain the quantization parameter that this coding unit should be taked according to preallocated code check afterwards.

The overall flow of Rate Control as shown in Figure 3.Rate Control is only at an image sets (between two I frames).For an image sets, control procedure is at first from initialization, comes bit to entire image set of dispense some according to target bit rate.A two field picture obtains target bit rate by code check preassignment in the image sets, obtains quantization parameter, the output bit flow of encoding afterwards by quantizing parameter Estimation.According to the bit of output and the difference between the target bit rate, correction image group overall bit rate is distributed, and then revises the code check preassignment of next frame.

Whole Rate Control process mainly is divided into following step:

1, according to the target bit rate requirement, the overall bit number of initialization present image group:

R＝bit_rate*L/F。Here bit_rate represents the target bit rate requirement, and L is the size (frame) of image sets, and F is a frame per second.Initialization destination buffer size, initialization histogram of difference average: ${\mathrm{hod}}_a=\mathrm{hod}\left(0\right)=\underset{i=0}{\overset{K=1}{\mathrm{\Σ}}}\mathrm{hod}(0,i),\mathrm{hod}(0,i)=\frac{\underset{\left|x\right|\≥a}{\mathrm{\Σ}}\mathrm{histogram}(0,i,x)}{N_p(0,i)}.$ Here hod represents histogram of difference frame (difference frame histogram), and hod (m) represents the frame difference histogram of m inter-frame of present image group, m=0 for initial inter-frame, and it also is current histogram mean value hod simultaneously _aHere suppose every frame K elementary cell, (m i) represents i elementary cell histogram of difference of m frame to hod.(m, i x) represent that difference value is the statistics with histogram value of x in the m frame i coding unit to Histogram.α is a threshold value, and value is between (0,20), and it has reflected the sensitivity of histogram for image difference, and α is more little, and sensitivity is high more.N _p(m i) is the number of pixels of present encoding unit.

2, to each frame, the encoder complexity of first coding unit of initialization $\mathrm{Xb}(m,0)=\frac{1}{\mathrm{Km}}\underset{k=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{i=0}{\overset{K-1}{\mathrm{\Σ}}}\mathrm{Xb}(k,i),$ m＞0；Xb(m，0)＝C，m＝0。Here Xb (k i) represents the encoder complexity of k inter-frame i coding unit, Xb (k, i)=r (k, i) * Q (k, i), r (k, i) the corresponding coding unit of the representative used bit of encoding, Q (k, i) the representative quantization parameter that corresponding coding unit adopted here.

3, calculate the histogram of difference of present frame and according to the mode preassignment code check of cumulative mean: $r\left(m\right)=(1+\frac{\mathrm{hod}\left(m\right)-\mathrm{ho}{d}_a}{{\mathrm{hod}}_a})\frac{R-R_u}{L-m-1},\mathrm{ho}{d}_a=\frac{1}{m+1}\underset{i=0}{\overset{m}{\mathrm{\Σ}}}\mathrm{hod}\left(i\right),$ Here R _uRepresent the code check that produced to preceding m+1 frame coding.Further combined with the destination buffer occupancy, be to the final code check preassignment of present frame:

r(m)＝λr(m)+(1-λ)(BL(m)-Tbl(m))。Here λ=0.5 is a weight coefficient, and BL (m) represents the occupancy of present encoding buffering area, and Tbl (m) takies the destination buffer occupancy that model obtains according to destination buffer.

4, distribute code check for the present encoding unit according to the histogram of difference of present encoding unit: $r(m,i)=(1+\frac{\mathrm{hod}(m,i)-\mathrm{ho}{d}_a}{{\mathrm{hod}}_a})\frac{r\left(m\right)-r_u}{K-i},$ Here ru is the encoded actual bit rate that elementary cell produced of present frame.

5, prediction present encoding cell encoding complexity.If first coding unit of present frame is then directly used step 2 initialization codes complexity.Otherwise,, also represented the own encoder complexity of image (motion Shaoxing opera is strong, and encoder complexity is big more) indirectly because histogram of difference itself has reflected the movement degree of image inside.So have: $\frac{\mathrm{Xb}(m,i)}{\mathrm{Xb}(m,i+1)}=\frac{\mathrm{hod}(m,i)}{\mathrm{hod}(m,i+1)}$ Such hypothesis.In addition, for video image, the continuity on the feature presentative time between the consecutive frame (feature that is consecutive frame has very big similitude) for this reason, can obtain following relation:

$-\mathrm{\ω}\≤\frac{\mathrm{hod}(m,i)}{\mathrm{hod}(m,i+1)}-\frac{\mathrm{hod}(m+1,i)}{\mathrm{hod}(m+1,i+1)}\≤\mathrm{\ω}$

Based on above-mentioned these two relations, can further provide following relation:

If $|\frac{\mathrm{hod}(m,i)}{\mathrm{hod}(m,i-1)}-\frac{\mathrm{hod}(m-1,i)}{\mathrm{hod}(m-1,i-1)}|\≤\mathrm{\ω}$ (condition 1) set up, then $\mathrm{Xb}(m,i)=\mathrm{Xb}(m,i-1)\frac{\mathrm{Xb}(m-1,i)}{\mathrm{Xb}(m-1,i-1)}$ (complexity method of estimation 1) ω here is a threshold value constant, and value is between [0.05,0.1].

If $|\frac{\mathrm{hod}(m,i)}{\mathrm{hod}(m,i-1)}-\frac{\mathrm{hod}(m-1,i)}{\mathrm{hod}(m-1,i-1)}|\≤\mathrm{\ω}$ Be false, then adopt following Forecasting Methodology (complexity method of estimation 2):

If | hod (m, i)-hod (m, i-1) |-| hod (m, i)-hod (m-1, i) |＞0.1hod _aSo Xb (m, i)=(2*Xb (and m-1, i)+Xb (m, i-1))/3 otherwise Xb (m, i)=(Xb (and m-1, i)+2*Xb (m, i-1))/3

6, quantization parameter calculates and regulates.Obtained the encoder complexity prediction of present encoding unit, the quantization parameter that the definition by encoder complexity obtains predicting: $Q(m,i)=\frac{\mathrm{Xb}(m,i)}{r(m,i)}.$ Because the code check r (m that the present encoding unit distributes, i) might be less than 0, but quantization parameter must require the quantization parameter of adjacent encoder unit can not differ too big greater than 0 and in order to guarantee the continuity of video pictures quality, so the quantization parameter that obtains also will be revised as follows:

Q(m，i)＝max(2，min(Q _a+3，max(Q _a-3，Q _a))，m＝0

Q (m, i)=max (2, and min (Q (m, i-1)+3, max (Q (m, i-1)-3, Q (m, i))), m＞0, min (.) and max (.) are illustrated respectively in two numbers and get less and higher value here, Q _aRepresent the mean value of encoded elementary cell quantization parameter.

Claims (4)

1, a kind of video code rate control method of low complex degree comprises bit rate pre-allocation process and quantization parameter adjustment process; It is characterized in that described bit rate pre-allocation process is to utilize video sequence characteristics to carry out code check preassignment; Described quantization parameter adjustment process is to come the predictive coding complexity according to video sequence characteristics, and then carries out quantization parameter and estimate; Described video sequence characteristics is meant the histogram of difference of representing the video sequence local motion.

2. by the video code rate control method of the described low complex degree of claim 1, it is characterized in that described code check preassignment comprises the steps:

Step 1): target setting code check and destination buffer occupancy and gop size; According to the histogram of difference of first each coding unit of inter-frame, the histogram of difference value of present frame is asked in initialization, and histogram mean value is initialized as this value; In same image sets, the mean value of successive image histogram of difference obtains according to the mode of cumulative mean;

Step 2): to each frame, the encoder complexity of first coding unit of initialization;

Step 3): calculate the difference of consecutive frame and add up its histogram, calculate the average of histogram of difference sequence; According to the difference between present frame histogram and the average, combining target code check and destination buffer occupancy are distributed frame level bit-rate according to average-scale model then;

Step 4): on the basis of frame one-level Data Rate Distribution,, adopt average one scale model to distribute the code check of basic cell level further according to the histogram of difference of each Rate Control elementary cell.

3. by the video code rate control method of the described low complex degree of claim 1, it is characterized in that described quantization parameter is estimated to comprise the steps:

Step 5): according to adjacent encoder unit histogram distribution in the frame, the encoder complexity of prediction present encoding unit;

Step 6): obtain after the encoder complexity of present encoding unit,, calculate the quantization parameter of present encoding unit according to preallocated code check; Utilize this quantization parameter that present image is encoded, and go to revise the Data Rate Distribution of next frame with the code check that obtains behind the coding.

4, by the video code rate control method of the described low complex degree of claim 2, it is characterized in that described Rate Control elementary cell is a macro-block line.

Images