CN102148995A - Adaptive intra-frame code rate prediction control algorithm based on rate distortion - Google Patents

Abstract

The invention relates to an adaptive intra-frame code rate prediction control algorithm based on rate distortion. In the algorithm, a novel code rate control model is provided in consideration of image complexity, cache state, scene change and other factors, thereby greatly improving encoding quality.

Description

Self-adaption intra-frame prediction frame bit rate control method based on rate distortion

Technical field

The present invention relates to field of video encoding, particularly a kind of adaptive infra-frame prediction frame Rate Control (Rate Control) algorithm based on rate distortion.

Background technology

H.264/AVC as the up-to-date video standard of ITU and MPEG joint development, obtained very big raising than traditional video standard.At present, existing a considerable amount of rate control algorithms propose in standard group, but their great majority all are at the P frame, are difficult to be directly applied for the I frame.In video coding system, be the key that encoder can satisfy the target bandwidth at last to the quality control of I frame.Because the bit number of I frame is more much bigger than same P frame or B frame, P frame, B frame can use the information of I frame to produce the image information of oneself simultaneously.Therefore, the quality decline of I frame can directly cause the quality of whole video sequence to descend.The present invention has considered factors such as image complexity, buffer status, scene change, proposes a kind of new Rate Control model, can improve coding quality significantly.

Summary of the invention

Because relation below the code check of I frame and MSE (MSE below appears being abbreviated as in mean square error) exist:

$\mathrm{BitRate}=\ln \left(\frac{1}{\mathrm{\λMSE}}\right)$

So algorithm of the present invention is not assert QP (QP below appears being abbreviated as in quantization parameter) and Qstep (Qstep below appears being abbreviated as in quantization step) and is linear relationship.

The objective of the invention is to realize by following steps:

1. the average gradient of computed image is weighed image complexity, and formula is as follows:

$\mathrm{Grad}=\frac{1}{\mathrm{MN}}\underset{i=1}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\left(\right|I_{i,j}-I_{i+1,j}|+|I_{i,j}-I_{i,j+1}\left|\right);---\left(1\right)$

I wherein _{I, j}For present frame (i, the brightness value of j) locating, M are picture traverse, N is a picture altitude.

2. obtain the MSE of first I frame of i GOP according to the JVT-G012 algorithm, be expressed as MSE ₁

3. according to test widely in a large number, find to exist between average gradient value and the output code flow pseudo-linear relation, according to the MSE of this Relationship Prediction present frame.It is as follows to be expressed as formula:

${\mathrm{MSE}}_i={\mathrm{MSE}}_{i-1}\×\frac{{\mathrm{Grad}}_i}{\mathrm{MG}}\×\frac{1}{1-{\mathrm{BR}}_{i-1}}+\mathrm{\θ};---\left(2\right)$

MSE wherein _iBe the I frame MSE value of i GOP (being current GOP), Grad _iBe the average gradient value of i GOP (being current GOP), MG is the average of the average gradient of the I frame of having encoded before this sequence, and θ is a constant, BR _iBe the memory space utilization rate of present encoding, formulate is as follows:

${\mathrm{BR}}_i=\frac{{\mathrm{BF}}_i}{\mathrm{Buffer}\_\mathrm{Size}};---\left(3\right)$

BF wherein _iFor behind i the GOP that encoded, the utilization rate of memory space.

4. according to test widely in a large number, the relation that draws between Qstep and the MSE can be expressed as following formula:

Qstep＝α×MSE+β； (4)

Wherein α can calculate after finishing first I frame coding, and according to the difference of video sequence and difference, β for same video sequence, is worth unique for being constant.

The generation of determining scene change according to DOH (DOH below appears being abbreviated as in image histogram) whether, formula is as follows:

${\mathrm{DOH}}_n=\frac{1}{M\×N}\underset{i=0}{\overset{255}{\mathrm{\Σ}}}|h_n\left(i\right)-h_{n-1}\left(i\right)|;---\left(5\right)$

If the DOH value greater than 0.5, is then thought scene change has taken place.

For algorithm performance of the present invention is described, oppose than test with dissimilar video sequences and JVT-G012 algorithm, be specially: " Akyio ", " Suzie ", " Silent ", " Paris ", " Paris-Silent ", " Paris-Foreman ", " Silent-Suzie ", " Paris-Suzie ", " Foreman ".

Table one

Can find out that from table 1 this compares with the JVT-G012 algorithm based on the self-adaption intra-frame prediction frame rate control algorithm of rate distortion, effect is obvious.

Specific implementation process

To this self-adaption intra-frame prediction frame rate control algorithm based on rate distortion, concrete implementation step is described below:

1. the average gradient of computed image is weighed image complexity, and formula is as follows:

$\mathrm{Grad}=\frac{1}{\mathrm{MN}}\underset{i=1}{\overset{M-1}{\mathrm{\Σ}}}\underset{j=1}{\overset{N-1}{\mathrm{\Σ}}}\left(\right|I_{i,j}-I_{i+1,j}|+|I_{i,j}-I_{i,j+1}\left|\right);---\left(1\right)$

I wherein _{I, j}For present frame (i, the brightness value of j) locating, M are picture traverse, N is a picture altitude.

2. obtain the MSE of first I frame of i GOP according to the JVT-G012 algorithm, be expressed as MSE ₁

3. according to test widely in a large number, find to exist between average gradient value and the output code flow pseudo-linear relation, according to the MSE of this Relationship Prediction present frame.It is as follows to be expressed as formula:

${\mathrm{MSE}}_i={\mathrm{MSE}}_{i-1}\×\frac{{\mathrm{Grad}}_i}{\mathrm{MG}}\×\frac{1}{1-{\mathrm{BR}}_{i-1}}+\mathrm{\θ};---\left(2\right)$

MSE wherein _iBe the I frame MSE value of i GOP (being current GOP), Grad _iBe the average gradient value of i GOP (being current GOP), MG is the average of the average gradient of the I frame of having encoded before this sequence, and θ is a constant, BR _iBe the memory space utilization rate of present encoding, formulate is as follows:

${\mathrm{BR}}_i=\frac{{\mathrm{BF}}_i}{\mathrm{Buffer}\_\mathrm{Size}};---\left(3\right)$

BF wherein _iFor behind i the GOP that encoded, the utilization rate of memory space.

4. according to test widely in a large number, the relation that draws between Qstep and the MSE can be expressed as following formula:

Qstep＝α×MSE+β； (4)

Wherein α can calculate after finishing first I frame coding, and according to the difference of video sequence and difference, β for same video sequence, is worth unique for being constant.

The generation of determining scene change according to DOH (DOH below appears being abbreviated as in image histogram) whether, formula is as follows:

${\mathrm{DOH}}_n=\frac{1}{M\×N}\underset{i=0}{\overset{255}{\mathrm{\Σ}}}|h_n\left(i\right)-h_{n-1}\left(i\right)|;---\left(5\right)$

If the DOH value greater than 0.5, is then thought scene change has taken place.

Claims (4)

1. the self-adaption intra-frame prediction frame rate control algorithm based on rate distortion is characterized in that, comprises the steps:

1) average gradient of computed image.

2) MSE of first I frame of i GOP of calculating.

3) according to the MSE of the Relationship Prediction present frame of average gradient value and output code flow.

4) parameter of linear relationship formula between calculating Qstep and the MSE.

5) calculate DOH, whether the generation of judgement scene change.

2. the self-adaption intra-frame prediction frame rate control algorithm based on rate distortion as claimed in claim 1 is characterized in that in the described step 3) formula of the MSE of prediction present frame:

${\mathrm{MSE}}_i={\mathrm{MSE}}_{i-1}\×\frac{{\mathrm{Grad}}_i}{\mathrm{MG}}\×\frac{1}{1-{\mathrm{BR}}_{i-1}}+\mathrm{\θ};---\left(2\right)$

MSE wherein _iBe the I frame MSE value of i GOP (being current GOP), Grad _iBe the average gradient value of i GOP (being current GOP), MG is the average of the average gradient of the I frame of having encoded before this sequence, and θ is a constant, BR _iBe the memory space utilization rate of present encoding, formulate is as follows:

${\mathrm{BR}}_i=\frac{{\mathrm{BF}}_i}{\mathrm{Buffer}\_\mathrm{Size}};---\left(3\right)$

BF wherein _iFor behind i the GOP that encoded, the utilization rate of memory space.

3. the self-adaption intra-frame prediction frame rate control algorithm based on rate distortion as claimed in claim 1 is characterized in that in the described step 4) linear relationship formula between the Qstep and MSE:

Qstep＝α×MSE+β； (4)

Wherein α can calculate after finishing first I frame coding, and according to the difference of video sequence and difference, β for same video sequence, is worth unique for being constant.

4. the self-adaption intra-frame prediction frame rate control algorithm based on rate distortion as claimed in claim 1, the generation whether principle that it is characterized in that calculating the formula of DOH in the described step 5) and judge scene change:

${\mathrm{DOH}}_n=\frac{1}{M\×N}\underset{i=0}{\overset{255}{\mathrm{\Σ}}}|h_n\left(i\right)-h_{n-1}\left(i\right)|;---\left(5\right)$

If the DOH value greater than 0.5, is then thought scene change has taken place.