CN102148973A - Three-layer rate control method based on Lagrange's multiplier factors - Google Patents

Abstract

The invention discloses a three-layer rate control method based on Lagrange's multiplier factors. At a macro block stage, a Lagrange's multiplier factor model is dynamically corrected in the algorithm, so that a preferable encoding mode is selected. At a frame stage, a quantitative parameter selection method which is more flexible and effective to I frame is used in the algorithm, thereby avoiding the overflow of an internal memory. At a group of picture (GOP) stage, an effective bit distribution method is provided in the algorithm.

Description

3 layer bit rate control methods based on the Lagrange's multiplier factor

Technical field

The present invention relates to field of video encoding, particularly a kind of 3 layer bit rates control (Rate Control) algorithm based on the Lagrange's multiplier factor.

Background technology

H.264/AVC the up-to-date video standard of developing jointly as ITU and MPEG has obtained very big raising than traditional video standard, and has integrated a lot of advanced technologies, is exactly wherein a kind of based on the rate-distortion optimization technology of Lagrange's equation.In H.264/AVC, the Lagrange's multiplier factor is a constant in macro-block level, though satisfied re-set target, for the frame condition of complexity, but is not optimal selection.Because the image in the real world is even same frame also has the different complexities and the distortion factor.

Summary of the invention

Purpose of the present invention comprises the Bit Allocation in Discrete of the adjustment, quantization parameter adjustment, GOP (GOP below appears being abbreviated as in set of pictures) of the self adaptation Lagrange's multiplier factor, is to realize by following steps:

1. determine image complexity, calculate the MAD (MAD below appears being abbreviated as in mean absolute difference) of i macro block of n-1 frame (former frame), formula is as follows:

${\mathrm{MAD}}_{\mathrm{MB}[i,n-1]}=\frac{1}{M\×N}\×\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}|I_{\mathrm{cur}}(x,y)-I_{\mathrm{ref}}(x-\mathrm{\Δx},y-\mathrm{\Δy})|;---\left(1\right)$

I wherein _Cur(x y) is positioned at (x, brightness value y), I for present frame _Ref(x-Δ x, y-Δ y) is positioned at the brightness value of (x-Δ x, y-Δ y) for reference frame, and Δ x and Δ y are motion vector, and M is the horizontal dimensions of current macro, and N is the vertical dimensions of current macro.

2. calculate the adjustment parameter of the Lagrange's multiplier factor, for i macro block of n frame (present frame),

Formula is as follows:

${\mathrm{\α}}_{i,n}={\left(\frac{{\mathrm{MAD}}_{\mathrm{Frame}[n-1]}}{{\mathrm{MAD}}_{\mathrm{MB}[i,n-1]}}\right)}^{\mathrm{\φ}};---\left(2\right)$

MAD wherein _Frame[n-1]Be the mean value of each macro block MAD of n-1 frame (former frame), MAD _{MB[i, n-1]}Be the MAD of n-1 frame (former frame) i macro block, parameter phi is chosen according to actual needs, and recommended value is 0.25.

3. calculate the Lagrange's multiplier factor of i macro block of n frame (present frame), formula is as follows:

λ _MODE[i，n]＝α _i，n×λ _MODE； (3)

λ _MODE＝0.85×2 ^(QP-12)/3； (4)

Wherein QP is a quantization parameter.

4. calculate I frame quantization parameter, formula is as follows:

${\mathrm{QP}}_I=\left(\begin{array}{c}{\mathrm{QP}}_I-1,\mathrm{BR}\≤{\mathrm{th}}_1\≤\frac{N_{\mathrm{SP}}}{N_p}\\ {\mathrm{QP}}_I+1,{\mathrm{th}}_1\≤\frac{N_{\mathrm{SP}}}{N_p}\≤{\mathrm{th}}_2\\ {\mathrm{QP}}_I+1,\mathrm{BR}\≥{\mathrm{th}}_2\\ {\mathrm{QP}}_I+2,{\mathrm{th}}_2\≤\frac{N_{\mathrm{SP}}}{N_p}\≤{\mathrm{th}}_3\\ {\mathrm{QP}}_I+4,\frac{N_{\mathrm{SP}}}{N_p}>{\mathrm{th}}_3\end{array}\right.;---\left(5\right)$

N wherein _SPBe the P frame number of skipping among the previous GOP, N _pBe P frame number total among the previous GOP.Th ₁, th ₂, th ₃3 valve values for the current cache utilance are set at 10%, 30%, 55% usually, and BR is the buffer memory occupancy before the current GOP of coding.

5. in whole cataloged procedure H.264/AVC, the Bit Allocation in Discrete of GOP level all determined by formula (6), and that current GOP former frame coding is finished the also remaining bit number in back is definite by formula (7).And in the present invention, when previous GOP had used too much bit, the Bit Allocation in Discrete of GOP level changes by formula (8) to be determined, also remaining bit number was determined by formula (10) after current GOP last frame coding was finished.Each formula is as follows:

$T_r\left(n_{i,0}\right)=\frac{u\left(n_{i,1}\right)}{F_r}\×N_{\mathrm{GOP}}+T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(6\right)$

N wherein _GOPBe a totalframes among the GOP, u (n _{I, 1}) when being first frame of i GOP, available bandwidth,

In the previous GOP process of encoding, the occupation rate of buffer memory, F _rBe predefined frame per second.

T _r(n _i，j)＝T _r(n _i，j-1)-A(n _i，j-1)； (7)

A (n wherein _{I, j-1}) be in i GOP, the bit number behind the j-1 frame coding, T _r(n _{I, j}) for after finishing at j-1 frame coding, also remaining bit number.

$T_r\left(n_{i,0}\right)=\frac{u\left(n_{i,1}\right)}{F_r}\×N_{\mathrm{GOP}}+\mathrm{\β}\×T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(8\right)$

β＝a×N _GOP+b； (9)

The present invention is after extensive testing, and a=0.003, b=0.02 are got in suggestion.

$\stackrel{\‾}{T_r}\left(n_{i,N_{\mathrm{GOP}}}\right)=T_r\left(n_{i,N_{\mathrm{GOP}}}\right)+(1-\mathrm{\β})\×T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(10\right)$

Wherein

For current GOP last frame coding finish after the correction value of remaining bit number also.

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

Table one

Can find out that from table 1 this compares with the JVT-G012 algorithm based on 3 layer bit rate control algolithms of the Lagrange's multiplier factor, effect is obvious.

Specific implementation process

To these 3 layer bit rate control algolithms based on the Lagrange's multiplier factor, concrete implementation step is described below:

1. determine image complexity, calculate the MAD (MAD below appears being abbreviated as in mean absolute difference) of i macro block of n-1 frame (former frame), formula is as follows:

${\mathrm{MAD}}_{\mathrm{MB}[i,n-1]}=\frac{1}{M\×N}\×\underset{x=0}{\overset{M-1}{\mathrm{\Σ}}}\underset{y=0}{\overset{N-1}{\mathrm{\Σ}}}|I_{\mathrm{cur}}(x,y)-I_{\mathrm{ref}}(x-\mathrm{\Δx},y-\mathrm{\Δy})|;---\left(1\right)$

I wherein _Cur(x y) is positioned at (x, brightness value y), I for present frame _Ref(x-Δ x, y-Δ y) is positioned at the brightness value of (x-Δ x, y-Δ y) for reference frame, and Δ x and Δ y are motion vector, and M is the horizontal dimensions of current macro, and N is the vertical dimensions of current macro.

2. calculate the adjustment parameter of the Lagrange's multiplier factor, for i macro block of n frame (present frame), formula is as follows:

${\mathrm{\α}}_{i,n}={\left(\frac{{\mathrm{MAD}}_{\mathrm{Frame}[n-1]}}{{\mathrm{MAD}}_{\mathrm{MB}[i,n-1]}}\right)}^{\mathrm{\φ}};---\left(2\right)$

MAD wherein _Frame[n-1]Be the mean value of each macro block MAD of n-1 frame (former frame), MAD _{MB[i, n-1]}Be the MAD of n-1 frame (former frame) i macro block, parameter phi is chosen according to actual needs, and recommended value is 0.25.

3. calculate the Lagrange's multiplier factor of i macro block of n frame (present frame), formula is as follows:

λ _MODE[i，n]＝α _i，n×λ _MODE； (3)

λ _MODE＝0.85×2 ^(QP-12)/3； (4)

Wherein QP is a quantization parameter.

4. calculate I frame quantization parameter, formula is as follows:

${\mathrm{QP}}_I=\left(\begin{array}{c}{\mathrm{QP}}_I-1,\mathrm{BR}\≤{\mathrm{th}}_1\≤\frac{N_{\mathrm{SP}}}{N_p}\\ {\mathrm{QP}}_I+1,{\mathrm{th}}_1\≤\frac{N_{\mathrm{SP}}}{N_p}\≤{\mathrm{th}}_2\\ {\mathrm{QP}}_I+1,\mathrm{BR}\≥{\mathrm{th}}_2\\ {\mathrm{QP}}_I+2,{\mathrm{th}}_2\≤\frac{N_{\mathrm{SP}}}{N_p}\≤{\mathrm{th}}_3\\ {\mathrm{QP}}_I+4,\frac{N_{\mathrm{SP}}}{N_p}>{\mathrm{th}}_3\end{array}\right.;---\left(5\right)$

N wherein _SPBe the P frame number of skipping among the previous GOP, N _pBe P frame number total among the previous GOP.Th ₁, th ₂, th ₃3 valve values for the current cache utilance are set at 10%, 30%, 55% usually, and BR is the buffer memory occupancy before the current GOP of coding.

5. in whole cataloged procedure H.264/AVC, the Bit Allocation in Discrete of GOP level all determined by formula (6), and that current GOP former frame coding is finished the also remaining bit number in back is definite by formula (7).And in the present invention, when previous GOP had used too much bit, the Bit Allocation in Discrete of GOP level changes by formula (8) to be determined, also remaining bit number was determined by formula (10) after current GOP last frame coding was finished.Each formula is as follows:

$T_r\left(n_{i,0}\right)=\frac{u\left(n_{i,1}\right)}{F_r}\×N_{\mathrm{GOP}}+T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(6\right)$

N wherein _GOPBe a totalframes among the GOP, u (n _{I, 1}) when being first frame of i GOP, available bandwidth,

In the previous GOP process of encoding, the occupation rate of buffer memory, F _rBe predefined frame per second.

T _r(n _i，j)＝T _r(n _i，j-1)-A(n _i，j-1)； (7)

A (n wherein _{I, j-1}) be in i GOP, the bit number behind the j-1 frame coding, T _r(n _{I, j}) for after finishing at j-1 frame coding, also remaining bit number.

$T_r\left(n_{i,0}\right)=\frac{u\left(n_{i,1}\right)}{F_r}\×N_{\mathrm{GOP}}+\mathrm{\β}\×T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(8\right)$

β＝a×N _GOP+b； (9)

The present invention is after extensive testing, and a=0.003, b=0.02 are got in suggestion.

$\stackrel{\‾}{T_r}\left(n_{i,N_{\mathrm{GOP}}}\right)=T_r\left(n_{i,N_{\mathrm{GOP}}}\right)+(1-\mathrm{\β})\×T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(10\right)$

Wherein

For current GOP last frame coding finish after the correction value of remaining bit number also.

Claims (6)

1. 3 layer bit rate control algolithms based on the Lagrange's multiplier factor is characterized in that, comprise the steps:

1) MAD of i macro block of calculating n-1 frame.

2) the adjustment parameter of the calculating Lagrange's multiplier factor.

3) the Lagrange's multiplier factor of i macro block of calculating n frame.

4) calculate I frame quantization parameter.

5) determine the Bit Allocation in Discrete of GOP level.

2. 3 layer bit rate control algolithms based on the Lagrange's multiplier factor as claimed in claim 1 is characterized in that described step 2) the middle formula that calculates the adjustment parameter of the Lagrange's multiplier factor:

${\mathrm{\α}}_{i,n}={\left(\frac{{\mathrm{MAD}}_{\mathrm{Frame}[n-1]}}{{\mathrm{MAD}}_{\mathrm{MB}[i,n-1]}}\right)}^{\mathrm{\φ}};---\left(2\right)$

MAD wherein _Frame[n-1]Be the mean value of each macro block MAD of n-1 frame (former frame), MAD _{MB[i, n-1]}Be the MAD of n-1 frame (former frame) i macro block, the parameter phi recommended value is 0.25.

3. 3 layer bit rate control algolithms based on the Lagrange's multiplier factor as claimed in claim 1 is characterized in that in the described step 3) calculating the formula of the Lagrange's multiplier factor of i macro block of n frame:

λ _MODE[i，n]＝α _i，n×λ _MODE； (3)

λ _MODE＝0.85×2 ^(QP-12)/3； (4)

Wherein QP is a quantization parameter.

4. 3 layer bit rate control algolithms based on the Lagrange's multiplier factor as claimed in claim 1 is characterized in that in the described step 4) calculating the formula of I frame quantization parameter:

${\mathrm{QP}}_I=\left(\begin{array}{c}{\mathrm{QP}}_I-1,\mathrm{BR}\≤{\mathrm{th}}_1\≤\frac{N_{\mathrm{SP}}}{N_P}\\ {\mathrm{QP}}_I+1,{\mathrm{th}}_1\≤\frac{N_{\mathrm{SP}}}{N_P}\≤{\mathrm{th}}_2\\ {\mathrm{QP}}_I+1,\mathrm{BR}\≥{\mathrm{th}}_2\\ {\mathrm{QP}}_I+2,{\mathrm{th}}_2\≤\frac{N_{\mathrm{SP}}}{N_P}\≤{\mathrm{th}}_3\\ {\mathrm{QP}}_I+4,\frac{Q_{\mathrm{SP}}}{N_P}>{\mathrm{th}}_3\end{array}\right.;---\left(5\right)$

N wherein _SPBe the P frame number of skipping among the previous GOP, N _PBe P frame number total among the previous GOP.Th ₁, th ₂, th ₃3 valve values for the current cache utilance are set at 10%, 30%, 55% usually, and BR is the buffer memory occupancy before the current GOP of coding.

5. 3 layer bit rate control algolithms based on the Lagrange's multiplier factor as claimed in claim 1 is characterized in that being different from the described step 5) formula of the Bit Allocation in Discrete of definite GOP level H.264/AVC:

$T_r(n_i,0)=\frac{u(n_i,1)}{F_r}\×F_{\mathrm{GOP}}+\mathrm{\β}\×T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(8\right)$

β＝a×N _GOP+b； (9)

$\stackrel{\‾}{T_r}\left(n_{i,N_{\mathrm{GOP}}}\right){=T}_r\left(n_{i,N_{\mathrm{GOP}}}\right)+(1-\mathrm{\β})\×T_r\left(n_{i-1,N_{\mathrm{GOP}}}\right);---\left(10\right)$

Wherein

For current GOP last frame coding finish after the correction value of remaining bit number also.

6. 3 layer bit rate control algolithms based on the Lagrange's multiplier factor as claimed in claim 1, it is characterized in that in the described step 5) determining the using priciple of formula of the Bit Allocation in Discrete of GOP level, wherein formula (6) and formula (7) do not belong to the feature of the described 3 layer bit rate control algolithms based on the Lagrange's multiplier factor of claim 1:

In whole cataloged procedure H.264/AVC, the Bit Allocation in Discrete of GOP level all determined by formula (6), and that current GOP former frame coding is finished the also remaining bit number in back is definite by formula (7).And in the present invention, when previous GOP had used too much bit, the Bit Allocation in Discrete of GOP level changes by formula (8) to be determined, also remaining bit number was determined by formula (10) after current GOP last frame coding was finished.