CN105163119A - Code rate control method based on space-time domain context and motion complexity - Google Patents

Abstract

The invention relates to a code rate control method based on space-time domain context and motion complexity. The method comprises the following steps of: based on a R-lambda model, performing the first-step updating on model parameters alpha and Beta of a current basic unit C0 before coding of the C0; coding the current basic unit C0; waiting for completion of coding all the basic units of a frame where the C0 is, and performing the second-step updating on the current basic unit; guiding target bits allocation of a basic unit layer with the model parameters subjected to the two steps of parameter updating; defining a relative motion complexity coefficient of the current basic unit C0 to represent motion intensity of different basic units; regulating a lambda value of the current basic unit with Wi, j of the current basic unit C0 to make the lambda value adaptive to the motion intensity of different basic units, thereby realizing the aim of allocating more bits to the basic units with higher motion intensity. The code rate control method based on space-time domain context and motion complexity can guarantee that the basic units with higher motion intensity can get more target bits, thereby facilitating improvement of quality of video.

Description

A kind of bit rate control method based on Space-time domain context and motion complexity

Technical field

The invention belongs to field of video encoding, relate to a kind of bit rate control method improving the accuracy of Rate Control particularly.

Background technology

Along with the development of computer network, the communication technology and multimedia technology, video related application becomes more and more extensive.But video data is often very huge, the storage of video and be transferred to existing storage and communication system brings huge challenge, must utilize video compression technology to reduce memory capacity and the transmission bandwidth of video.And the important step of rate control techniques just in video compression technology, by Rate Control, under the constraint of code check and buffering area, by Automatic adjusument coding parameter, code stream and channel transmission bandwidth can be adapted, and make video distortion minimum.

The video encoding standard of a new generation, high-performance video coding (HEVC) standard, since issue, has mainly been proposed two kinds of bit rate control methods: based on the bit rate control method of URQ model and the bit rate control method based on R-λ model.URQ model description be the relation of target bits based on pixel and QP value, its advantage is that URQ model is Pixel-level, so all applicable for the elementary cell of different size.But the model due to the method still adopts QP as the factor of most important adjustment bit rate control method, and therefore Rate Control accuracy is not high.Compared to the bit rate control method based on URQ model, the bit rate control method based on R-λ model adopts λ domain model, and algorithm performance is better.But also there is certain limitation based on the bit rate control method of R-λ model, in parameter renewal, only make use of temporal correlation and ignore spatial coherence, make the parameter after renewal and inaccuracy; In elementary cell layer target bit allocation, based on following hypothesis: the elementary cell of same rank same position may have identical model parameter, if but current basic unit motion intense, then it differs usually comparatively large with the elementary cell of reference picture correspondence position, therefore both use identical model parameter to be inaccurate.

Summary of the invention

The present invention is intended to the above-mentioned deficiency improving prior art, provide a kind of model parameter more accurately, target bit allocation more rationally thus the bit rate control method making video quality more stable.In order to achieve the above object, the present invention adopts following technical scheme:

Based on a bit rate control method for Space-time domain context and motion complexity, comprise the following steps:

1) based on R-λ model, definition current basic unit C ₀adjacent 8 elementary cells as a context, be designated as C ₁~ C ₈; At C ₀before coding, first time renewal is carried out to its model parameter α and β; C ₀context in C ₁~ C ₄encoded, therefore there are four kinds of parameter prediction direction;

2) calculate the MAD distance on four parameter prediction directions respectively, select from four MAD values two minimum, the elementary cell in this both direction is and C ₀two elementary cells that spatial coherence is the highest; Thus the actual parameter of the encoded elementary cell in both direction is added C ₀model parameter upgrade in, complete first time parameter upgrade;

3) to current basic unit C ₀encode;

4) C is waited for ₀all elementary cell codings of place frame are complete, start to upgrade the second step parameter of current basic unit; The context of current basic unit completes coding all, therefore its 8 context C ₁~ C ₈all need to calculate and C ₀mAD distance, therefrom select two minimum elementary cells, use the model parameter of its actual coding to current basic unit C ₀carry out the renewal of second step parameter; Model parameter after renewal is wherein i and j represents current basic unit C ₀for i-th elementary cell of jth width picture;

5) model parameter after two step parameters renewals, will be used to guide the target bit allocation of elementary cell layer; Definition current basic unit C ₀relative motion complexity coefficient W _i,jcharacterize the motion intense degree of different elementary cell, W _i,jcalculation procedure is as follows:

(1) brightness absolute difference and average is adopted to calculate current basic unit C ₀motion complexity, formula is as follows:

${\mathrm{Diff}}_{(i,j)}=\frac{1}{M_{i,j}\·N_{i,j}}\underset{m=1}{\overset{M_{i,j}}{\Σ}}\underset{n=1}{\overset{N_{i,j}}{\Σ}}|I_j(m,n)-I_{j,ref}(m,n)|---\left(1\right)$

In formula, Diff _{(i, j)}represent absolute difference and the average of i-th the elementary cell brightness of jth width picture, M _i,jand N _i,jrepresent row pixel count and the row pixel count of jth width picture i-th elementary cell, represent in this elementary cell with (m, n) and be no more than M _i,j, N _i,ja pixel at place, I _j(m, n) and I _{j, ref}(m, n) represents jth frame and reference frame position (m, n) place pixel brightness value respectively;

(2) jth width picture mean motion complexity is calculated concrete formula is:

${\mathrm{Diff}}_j^{avg}=\frac{1}{M_j\·N_j}\underset{m=1}{\overset{M_j}{\Σ}}\underset{n=1}{\overset{N_j}{\Σ}}|I_j(m,n)-I_{j,ref}(m,n)|---\left(2\right)$

Wherein, M _jand N _jbe respectively the capable total pixel number of jth width picture and row total pixel number;

(3) predetermined threshold value Th, calculates the relative motion complexity coefficient W of current basic unit _i,j, computing formula is as follows:

$W_{i,j}=\left\{\begin{array}{cc}\frac{{\mathrm{Diff}}_{i,j}}{{\mathrm{Diff}}_j^{avg}},& \begin{array}{cc}if& {\mathrm{Diff}}_{i,j}\≥Th\end{array}\\ 1,& else\end{array}\right.---\left(3\right)$

$\mathrm{Th}=\mathrm{\δD}{\mathrm{iff}}_j^{\mathrm{avg}}$ Wherein δ gets 1.3;

6) current basic unit C is used ₀w _i,jregulate the λ value of current basic unit, make λ value adapt to the motion intense degree of different elementary cell, thus be embodied as the target of the more violent more bits of elementary cell distribution of motion; The λ value λ of jth width picture i-th BU _i,jbe defined as:

${\mathrm{\λ}}_{i,j}=\frac{{\mathrm{\λ}}_j}{W_{i,j}}---\left(4\right)$

Wherein, λ _jrepresent the λ value of jth width picture;

Then elementary cell target bits computing formula be improved to:

$T_{i,j}^w={\mathrm{\α}}_{i,j}^{new}\·{{\mathrm{\λ}}_{i,j}}^{{\mathrm{\β}}_{i,j}^{new}}---\left(5\right)$

Model parameter wherein use step 4) in the parameter that obtains of two step parameter updating methods.

The target bit allocation algorithm that the present invention proposes, can ensure that the basic unit block that motion complexity is higher assigns to more target bits, thus is conducive to the quality improving video.

Compared with prior art, this method has following several obvious advantages:

(1) video sequence not only interframe there is plenty of time redundancy, also there is spatial redundancy in frame, make full use of spatial coherence, add the adaptability of model parameter to video sequence, also can upgrade more accurately model parameter.

(2) by calculating the relative motion complexity coefficient of elementary cell, realizing the higher basic unit block of motion complexity and assigning to more target bits, improve the quality of video.

(3) parameter that the present invention proposes upgrades the innovatory algorithm with target bit allocation, and both are interactional.Parameter after renewal can be directly used in target bit allocation, and target bit allocation affects coding thus also has impact to model parameter, therefore parameter and target bit can make overall Rate Control more accurate more accurately, reduce fluctuation, make video quality more stable.

Accompanying drawing explanation

Fig. 1 is the flow chart of two step parameter updating methods in the present invention.

Fig. 2 is eight context schematic diagrames of the current basic unit periphery defined in the present invention.

Fig. 3 is the elementary cell layer target bit allocation algorithm flow chart based on motion complexity in the present invention.

Fig. 4 is that the PSNR of the innovatory algorithm that the HM11.0 algorithm of BQSquare sequence under target bits 1500Kbp proposes with the present invention when RA (the main class of random intervention) encodes compares.

Fig. 5 is that the PSNR of LB (the main class of low delay B) innovatory algorithm that the HM11.0 algorithm of BQSquare sequence under target bits 2000Kbps proposes with the present invention when encoding compares.

As can be seen from Fig. 4 and Fig. 5, the overall fluctuation of the Data Rate Distribution algorithm that the present invention proposes is all less, and video quality stability is higher, and video is more level and smooth.

Embodiment

For making object of the present invention, implementation and advantage more clear, below specific embodiment of the invention is described in further detail.

The present invention is to based on R-λ model ^[1]the improvement of bit rate control method, modified hydrothermal process can be integrated in reference software HM11.0, and specific implementation comprises new model parameter and upgrades and elementary cell layer target bit allocation two parts.

The concrete implementation step of two step parameter updating methods that the present invention proposes is as follows:

(1) current basic unit is C ₀, its periphery 8 elementary cells, as a context, are designated as C ₁~ C ₈.

(2) at C ₀before coding, C ₁~ C ₄encoded complete, therefore there is four kinds of parameter Forecasting Methodologies, i.e. horizontal direction, vertical direction, direction, the upper right corner and direction, the upper left corner.Because the elementary cell of left vicinity is the most relevant to active cell, therefore C ₀prediction direction be utilize C ₄and C ₄context to determine.

(3) calculate the MAD distance in four prediction direction, formula is as follows:

MAD _V＝|MAD _act[C ₁]-MAD _act[C ₄]|

MAD _H＝|MAD _act[C ₂]-MAD _act[C ₁]|

MAD _R＝|MAD _act[C ₄]-MAD _act[C ₂]|

MAD _L＝|MAD _act[C ₄]-MAD _act[C ₁₀]|(9)

Wherein, MAD _v, MAD _h, MAD _r, MAD _lrepresent vertical direction respectively, horizontal direction, direction, the upper right corner and direction, the upper left corner.

(4) choosing the less both direction of MAD value is parameter prediction direction, such as MAD _v, MAD _hless, then select C ₀horizontal direction and vertical direction on elementary cell C ₄and C ₂carry out parameter renewal.Computational methods are as follows:

The actual bpp average of the elementary cell of two participation undated parameters:

${\mathrm{bpp}}_{ave}=\frac{{\mathrm{bpp}}_1+{\mathrm{bpp}}_2}{2}---\left(10\right)$

Corresponding actual λ _realaverage λ _avebe defined as:

${\mathrm{\λ}}_{ave}=\sqrt{{\mathrm{\λ}}_1\·{\mathrm{\λ}}_2}---\left(11\right)$

So λ _compbe redefined for computing formula is as follows:

${\mathrm{\λ}}_{comp}^{avg}={\mathrm{\α}}_{old}\·{{\mathrm{bpp}}_{ave}}^{{\mathrm{\β}}_{old}}---\left(12\right)$

Then before current BU is encoded, its model parameter upgrades as follows:

${\mathrm{\α}}_{new}^{\′}\≈{\mathrm{\α}}_{old}+{\mathrm{\δ}}_{\mathrm{\α}}({\mathrm{ln\λ}}_{ave}-{\mathrm{ln\λ}}_{comp}^{avg})\·{\mathrm{\α}}_{old}---\left(13\right)$

${\mathrm{\β}}_{new}^{\′}\≈{\mathrm{\β}}_{old}+{\mathrm{\δ}}_{\mathrm{\β}}({\mathrm{ln\λ}}_{ave}-{\mathrm{ln\λ}}_{comp}^{avg})\·{\mathrm{lnbpp}}_{ave}---\left(14\right)$

Wherein α ' _newwith β ' _newit is new model parameter.

(5) current basic unit is encoded, after waiting for that all elementary cells of present frame are all encoded, start second step parameter and upgrade.For C ₀, its context C1 ~ C8 is encoded complete, therefore eight elementary cells all can participate in parameter renewal.Calculate C ₀mAD difference in context, and the elementary cell that therefrom selection two MAD values are minimum carries out parameter renewal.MAD mathematic interpolation formula is as follows:

MAD _i＝|MAD _act[C _i]-MAD _act[C ₀]|i＝1,2,3,...,8(15)

(6) method be averaged is adopted to calculate new bpp _wand λ _w:

${\mathrm{bpp}}_w=\frac{{\mathrm{bpp}}_0\·w_0+{\mathrm{bpp}}_1\·w_1+{\mathrm{bpp}}_2\·w_2}{w_0+w_1+w_2}---\left(16\right)$

${\mathrm{\λ}}_w={({{\mathrm{\λ}}_0}^{w_0}\·{{\mathrm{\λ}}_1}^{w_1}\·{{\mathrm{\λ}}_2}^{w_2})}^{\frac{1}{w_0+w_1+w_2}}---\left(17\right)$

In formula, bpp ₀the actual bpp value of current basic unit, bpp ₁and bpp ₂participate in the actual bpp value that parameter upgrades contiguous BU.W ₀, w ₁, w ₂be set to 1,0.5 and 0.5 respectively.

(7) second step parameter updating method is as follows:

${\mathrm{\λ}}_{comp}^{avg}={\mathrm{\α}}_{old}\·{{\mathrm{bpp}}_{ave}}^{{\mathrm{\β}}_{old}}---\left(18\right)$

${\mathrm{\α}}_{new}\≈{\mathrm{\α}}_{old}+{\mathrm{\δ}}_{\mathrm{\α}}({\mathrm{ln\λ}}_w-{\mathrm{ln\λ}}_{comp}^w)\·{\mathrm{\α}}_{old}---\left(19\right)$

${\mathrm{\β}}_{new}\≈{\mathrm{\β}}_{old}+{\mathrm{\δ}}_{\mathrm{\β}}({\mathrm{ln\λ}}_w-{\mathrm{ln\λ}}_{comp}^w)\·\ln {\mathrm{bpp}}_w---\left(20\right)$

To sum up, two step parameter updating methods in the present invention are completed.

The concrete implementation step of elementary cell layer target bit allocation algorithm based on motion complexity that the present invention proposes is as follows:

(1) the brightness absolute difference and the average that calculate current basic unit characterize its motion complexity, and concrete formula is as follows:

${\mathrm{Diff}}_{(i,j)}=\frac{1}{M_{i,j}\·N_{i,j}}\underset{m=1}{\overset{M_{i,j}}{\Σ}}\underset{n=1}{\overset{N_{i,j}}{\Σ}}|I_j(m,n)-I_{j,ref}(m,n)|---\left(1\right)$

In formula, Diff _{(i, j)}represent absolute difference and the average of i-th the elementary cell brightness of jth width picture, M _i,jand N _i,jrepresent row pixel count and the row pixel count of jth width picture i-th elementary cell, represent in this elementary cell with (m, n) and be no more than M _i,j, N _i,ja pixel at place, I _j(m, n) and I _{j, ref}(m, n) represents jth frame and reference frame position (m, n) place pixel brightness value respectively.

(2) calculate the mean motion complexity of current basic unit place Picture, concrete formula is as follows:

${\mathrm{Diff}}_j^{avg}=\frac{1}{M_j\·N_j}\underset{m=1}{\overset{M_j}{\Σ}}\underset{n=1}{\overset{N_j}{\Σ}}|I_j(m,n)-I_{j,ref}(m,n)|---\left(2\right)$

Wherein, M _jand N _jbe respectively the capable total pixel number of jth width picture and row total pixel number, other parameters define in same step (1) and describe.

(3) calculate the relative motion complexity coefficient of current basic unit, be defined as follows:

$W_{i,j}=\left\{\begin{array}{cc}\frac{{\mathrm{Diff}}_{i,j}}{{\mathrm{Diff}}_j^{avg}},& \begin{array}{cc}if& {\mathrm{Diff}}_{i,j}\≥Th\end{array}\\ 1,& else\end{array}\right.---\left(3\right)$

$\mathrm{Th}={\mathrm{\δDiff}}_j^{\mathrm{avg}}$ Wherein δ gets 1.3

(4) according to the relative motion complexity coefficient of current basic unit, λ value is regulated, thus regulate target bits, make the strong elementary cell of motion Shaoxing opera be assigned to more target bit.Concrete formula is as follows:

λ value (the λ of jth width picture i-th BU _i,j) be defined as:

${\mathrm{\λ}}_{i,j}=\frac{{\mathrm{\λ}}_j}{W_{i,j}}---\left(4\right)$

Then formula is improved to:

$T_{i,j}^w={\mathrm{\α}}_{i,j}^{new}\·{{\mathrm{\λ}}_{i,j}}^{{\mathrm{\β}}_{i,j}^{new}}---\left(5\right)$

Model parameter wherein use the parameter that two step parameter updating methods proposed by the invention obtain.

New target bits computational methods are as follows:

${\tilde{T}}_{i,j}^w=\frac{T_r}{\underset{N_L}{\Σ}{W}_{i,j}}\×W_{i,j}---\left(6\right)$

$T_{i,j}=\frac{T_r}{\underset{N_L}{\Σ}{\hat{T}}_{i,j}^w}\×{\hat{T}}_{i,j}^w---\left(7\right)$

${\hat{T}}_{i,j}^w=\mathrm{\γ}\×{\tilde{T}}_{i,j}^w+(1-\mathrm{\γ})\×T_{i,j}^w---\left(8\right)$

In formula, be new target bit allocation proportion, the γ of Level1 (one-level) and Level2 (secondary) picture is set as 0.5, and other are set as 0.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, all any amendments done within the spirit and principles in the present invention, equivalent replacement and improvement etc., all should be included within protection scope of the present invention.

List of references

[1] Li Bin. towards the rate-distortion optimization technical research [D] of high-performance video coding standard. China Science & Technology University, 2013.DOI:10.7666/d.Y2354181.

Claims (1)

1., based on a bit rate control method for Space-time domain context and motion complexity, comprise the following steps:

1) based on R-λ model, definition current basic unit C ₀adjacent 8 elementary cells as a context, be designated as C ₁~ C ₈; At C ₀before coding, first time renewal is carried out to its model parameter α and β; C ₀context in C ₁~ C ₄encoded, therefore there are four kinds of parameter prediction direction;

2) calculate the MAD distance on four parameter prediction directions respectively, select from four MAD values two minimum, the elementary cell in this both direction is and C ₀two elementary cells that spatial coherence is the highest; Thus the actual parameter of the encoded elementary cell in both direction is added C ₀model parameter upgrade in, complete first time parameter upgrade;

3) to current basic unit C ₀encode;

4) C is waited for ₀all elementary cell codings of place frame are complete, start to upgrade the second step parameter of current basic unit: the context of current basic unit completes coding all, therefore its 8 context C ₁~ C ₈all need to calculate and C ₀mAD distance, therefrom select two minimum elementary cells, use the model parameter of its actual coding to current basic unit C ₀carry out the renewal of second step parameter; Model parameter after renewal is wherein i and j represents current basic unit C ₀for i-th elementary cell of jth width picture;

5) model parameter after two step parameters renewals, is used to guide the target bit allocation of elementary cell layer, definition current basic unit C ₀relative motion complexity coefficient W _i,jcharacterize the motion intense degree of different elementary cell, W _i,jcalculation procedure is as follows:

(1) brightness absolute difference and average is adopted to calculate current basic unit C ₀motion complexity, formula is as follows:

${\mathrm{Diff}}_{(i,j)}=\frac{1}{M_{i,j}\·N_{i,j}}\underset{m=1}{\overset{M_{i,j}}{\Σ}}\underset{n=1}{\overset{N_{i,j}}{\Σ}}|I_j(m,n)-I_{j,ref}(m,n)|---\left(1\right)$

In formula, Diff _{(i, j)}represent absolute difference and the average of i-th the elementary cell brightness of jth width picture, M _i,jand N _i,jrepresent row pixel count and the row pixel count of jth width picture i-th elementary cell, represent in this elementary cell with (m, n) and be no more than M _i,j, N _i,ja pixel at place, I _j(m, n) and I _{j, ref}(m, n) represents jth frame and reference frame position (m, n) place pixel brightness value respectively;

(2) jth width picture mean motion complexity is calculated concrete formula is:

${\mathrm{Diff}}_j^{avg}=\frac{1}{M_j\·N_j}\underset{m=1}{\overset{M_j}{\Σ}}\underset{n=1}{\overset{N_j}{\Σ}}|I_j(m,n)-I_{j,ref}(m,n)|---\left(2\right)$

Wherein, M _jand N _jbe respectively the capable total pixel number of jth width picture and row total pixel number;

(3) predetermined threshold value Th, calculates the relative motion complexity coefficient W of current basic unit _i,j, computing formula is as follows:

$W_{i,j}=\left\{\begin{array}{cc}\frac{{\mathrm{Diff}}_{i,j}}{{\mathrm{Diff}}_j^{avg}},& \begin{array}{cc}if& {\mathrm{Diff}}_{i,j}\≥Th\end{array}\\ 1,& else\end{array}\right.---\left(3\right)$

$Th={\mathrm{\δDiff}}_j^{avg}$ Wherein δ gets 1.3;

6) current basic unit C is used ₀w _i,jregulate the λ value of current basic unit, make λ value adapt to the motion intense degree of different elementary cell, thus be embodied as the target of the more violent more bits of elementary cell distribution of motion: the λ value λ of jth width picture i-th BU _i,jbe defined as:

${\mathrm{\λ}}_{i,j}=\frac{{\mathrm{\λ}}_j}{W_{i,j}}---\left(4\right)$

Wherein, λ _jrepresent the λ value of jth width picture;

Then elementary cell target bits computing formula be improved to:

$T_{i,j}^w={\mathrm{\α}}_{i,j}^{new}\·{{\mathrm{\λ}}_{i,j}}^{{\mathrm{\β}}_{i,j}^{new}}---\left(5\right)$

Model parameter wherein use step 4) in the parameter that obtains of two step parameter updating methods.