CN105376571A - HEVC reconstruction video scintillation inhibition method based on R-lambda code rate control model - Google Patents

Abstract

The invention discloses a HEVC (high efficiency video coding) reconstruction video scintillation inhibition method based on an R-lambda code rate control model. According to the method, motion information and complexity information of a CTU are compared with global motion information and complexity information of a present I frame, a type of the present CTU is determined, e.g., a motion CTU, a complex texture CTU or a flat texture CTU), a target code rate is distributed for CTU optimization according to the types of the CTU and residual target code rates, a QP value of the CTU is calculated by utilizing the R-lambda code rate control model, and reconstruction video scintillation inhibition is realized in a mode of code rate control. Through the HEVC reconstruction video scintillation inhibition method based on the R-lambda code rate control model, on the condition that video coding quality is guaranteed, a reconstruction video scintillation phenomenon is substantially inhibited, and the reconstruction video watch effect is improved.

Description

A kind of HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model

Technical field

The invention belongs to multimedia coding field, be specifically related to a kind of HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model.

Background technology

Along with network technology and computer process ability improve constantly, people propose new requirement to existing video encoding standard, wish high definition, ultra high-definition resolution video can be provided to compress, to meet the application in the fields such as new home theater, remote monitoring, digital broadcasting, mobile flow medium, mobile imaging and medical imaging.For this reason, JCT-VC (JointCollaborativeTeamonVideoCoding) has formally issued video encoding standard HEVC of new generation (HighEfficiencyVideoCoding) in January, 2013.

The target of HEVC is on basis H.264/AVChighprofile, dividing by adopting quadtree coding block more flexibly, predicting based on infra-frame prediction and type of prediction, the adaptive motion vector in direction and selecting the new technologies such as machine-processed, under the prerequisite ensureing same video picture quality, the code check of video flowing reduces 50%, and namely compression efficiency doubles.But HEVC still adopts and H.264/AVC similar coding structure, and intracoded frame (I frame) just becomes one of key factor affecting reconstruction video quality.In video coding process, in order to fault-tolerant, scene switches and the needs such as arbitrary access, usually in encoded video sequence, periodically insert I frame.But because the code efficiency of I frame is far below inter-frame encoding frame (P frame, B frame), when target bit rate is not enough, if do not controlled I frame in an encoding process, the I frame quality after rebuilding will be caused lower than adjacent P frame or B frame, video " flicker " phenomenon is just there will be, as shown in Figure 1 in reconstruction video playing process.Fig. 1 is the reconstruction image after normal video cycle tests Container encodes, wherein, 96th frame and the 98th frame are P frame, 97th frame is I frame, because the code efficiency of I frame is low, reconstruction quality is poor, causes occurring that between consecutive frame Subjective video quality is unsmooth, can feel that when people watch video constantly " flicker ", has a strong impact on video-see quality.

Rate control module in Video coding calculates QP value according to given target bit rate, and QP value directly has influence on the reconstruction quality of video, plays an important role in Video coding.In the identifying code HM10.0 of HEVC, provide two kinds of Rate Control reference models altogether, a kind of is URQ (UnifiedRate-Quantization) model based on pixel, and one is R-lambda model.URQ model can play good control effects to the encoding block of varying dimensions in HEVC, but in the Data Rate Distribution stage, URQ model is not distinguished I frame, does not consider the reconstruction quality improving I frame.Therefore, when adopting URQ model based coding, video " flicker " phenomenon that periodicity insertion I frame causes is comparatively serious, and this model is replaced by R-lambda model gradually.R-lambda model considers the different code efficiency of all types of frame, when distributing target bit rate for I frame, can adjust, for I frame distributes more target bit rate according to bpp (bitsperpixel) value of I frame.Adjustment mode is as shown in table 1, wherein, and T _curPicbe expressed as the target bit rate that I frame distributes.

I frame target bit rate method of adjustment in table 1R-lambda Rate Control model

bpp	bpp>0.2	0.2≥bpp>0.1	Other
				T after adjustment CurPic	5×T CurPic	7×T CurPic	10×T CurPic

But this adjustment underaction, can not change along with the change of video content.Through adjustment, although the reconstruction video quality of I frame is improved, the target bit rate consumed due to I frame is too much, easily causes P frame/B frame target bit rate thereafter not enough, affects its reconstruction quality.Especially last some frames of a video sequence, reconstruction quality declines serious, as shown in Figure 2.Fig. 2 (a) and Fig. 2 (b) respectively illustrates in HM10.0, adopting the reconstruction video objective quality (PSNR) after R-lambda Rate Control model based coding standard test sequences Container (352 × 288) and Boat (704 × 576), all there is larger decline in the reconstruction quality of two last some frames of sequence.But if the target bit rate distributing to I frame is not enough, because the code efficiency of I frame is lower, after several CTU that encoded, target bit rate will approach exhaustion, as shown in Figure 3.Follow-up CTU encodes owing to cannot obtain enough target bit rates, and its QP value can increase gradually, causes the reconstruction quality of CTU seriously to glide, and affects the whole body reconstruction quality to I frame.The present invention is from the angle of Rate Control, target bit rate is distributed by optimizing, reach the subjective quality utilized between the level and smooth I frame of less code check and consecutive frame, suppress the object of video " flicker ", simultaneously, improve the overall objective quality of reconstruction video, avoid the situation occurring that the last some frame reconstruction qualities of video sequence decline.

Summary of the invention:

The technical problem to be solved in the present invention is to provide a kind of HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model, obviously suppresses reconstruction video " flicker " phenomenon, improves reconstruction video viewing effect.

The present invention adopts following technical scheme to realize goal of the invention:

Based on a HEVC reconstruction video flicker restraint method for R-lambda Rate Control model, it is characterized in that, the step of the method is as follows:

(1), for intracoded frame, i.e. I frame, predicts movable information and the complexity information of current code tree unit CTU to be encoded, and the whole frame complexity information of current I frame;

(2), according to the movable information of CTU, extract the global motion information of I frame, and judge whether current C TU is motion CTU;

(3), for non-athletic CTU, according to the complexity information of current C TU and the complexity information of current I frame, judge that current C TU is the complicated CTU of texture, or the smooth CTU of texture;

(4), according to different CTU types, distribute target bit rate for CTU optimizes, and adopt R-lambda model to calculate the quasi-peak value of CTU, i.e. QP value.

As the further restriction to the technical program, described step (1) specifically comprises the steps:

(11), obtain the CTU of current CTU to be encoded in former frame corresponding position, extract its movable information;

(12), linear prediction method is adopted, the movable information of prediction current C TU;

(13), obtain the whole frame complexity information of the former frame of I frame, and current CTU to be encoded is in the complexity information of former frame corresponding position CTU;

(14), linear prediction method is adopted, the whole frame complexity information of prediction current I frame and the complexity information of current C TU.

As the further restriction to the technical program, described step (2) specifically comprises the steps:

(21), the motion vector of each CTU of I frame is decomposed into horizontal motion vector MV _xwith vertical motion vectors MV _y;

(22) MV of each CTU, is added up respectively _xand MV _y, calculate the MV occurring maximum probability _xand MV _y, it can be used as the horizontal direction of I frame and the global motion vector MV of vertical direction _gxand MV _gy;

(23), by the MV of each CTU _xand MV _ywith MV _gxand MV _gycompare, judge whether current C TU is motion CTU.

As the further restriction to the technical program, described step (4) specifically comprises the steps:

(41) the residue target bit rate of current I frame, is calculated;

(42), when remaining target bit rate and being greater than 0, being motion CTU distribution target bit rate according to the motion vector value of CTU and the motion vector value of I frame, according to the complexity information of CTU and the complexity information of I frame, is that non-athletic CTU distributes target bit rate;

(43), when remaining code check and being less than or equal to 0, calculating the actual coding code check average of current encoded CTU, is that current C TU distributes target bit rate according to this average.

As the further restriction to the technical program, in described step (42), the concrete grammar distributing target bit rate for motion CTU is:

$T_i\left(j\right)=\left\{\begin{array}{cc}\frac{{\mathrm{MV}}_{Gx}}{{\mathrm{MV}}_x}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}& {\mathrm{MV}}_x\&GreaterEqual;{\mathrm{MV}}_y\\ \frac{{\mathrm{MV}}_{Gy}}{{\mathrm{MV}}_y}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}& {\mathrm{MV}}_x<{\mathrm{MV}}_y\end{array}\right.$

The concrete grammar distributing target bit rate for the complicated CTU of texture and the smooth CTU of texture is:

$T_i\left(j\right)=\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTU}}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}$

Wherein, MV _gxfor the component of the global motion vector (x-axis direction) in the horizontal direction of current I frame, MV _xfor the component of the motion vector (x-axis direction) in the horizontal direction of current C TU, T _ifor the target bit rate that i-th I frame distributes, T _codedCTUfor the actual coding code check sum of CTU encoded in current I frame, N _left: the number of uncoded CTU in i-th I frame, MV _yfor the component of the motion vector (y-axis direction) in the vertical direction of current C TU, MV _gyfor the component of the global motion vector (y-axis direction) in the vertical direction of current I frame, MAD _intraFramefor the whole two field picture complexity of current I frame, MAD _curCTUfor the complexity of current C TU.

As the further restriction to the technical program, in described step (43), the concrete grammar distributing target bit rate for motion CTU and the complicated CTU of texture is:

$T_i\left(j\right)=\frac{1}{j-1}\&CenterDot;T_{codedCTN}$

The concrete grammar distributing target bit rate for the smooth CTU of texture is:

$T_i\left(j\right)=\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTN}}\&CenterDot;\frac{1}{j-1}\&CenterDot;T_{codedCTU},$

Wherein, T _ij () is the target bit rate of a jth CTU in i-th I frame, i and j is positive integer, T _codedCTUfor the actual coding code check sum of CTU encoded in current I frame, MAD _intraFramefor the whole two field picture complexity of current I frame, MAD _curCTUfor the complexity of current C TU.

Compared with prior art, advantage of the present invention and good effect are: the present invention proposes a kind of HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model, first the global motion information of the movable information of CTU and complexity information and current I frame and complexity information contrast by the method, judge type (the motion CTU of current C TU, the complicated CTU or the smooth CTU of texture of texture), then according to the type of CTU and I frame residue target bit rate for this CTU optimizes distribution target bit rate, R-lambda model is utilized to calculate the QP value of this CTU, realize video " flicker " in the angle of Rate Control to suppress.The present invention under the prerequisite ensureing video encoding quality, obviously can suppress reconstruction video " flicker " phenomenon, improves reconstruction video viewing effect.

Accompanying drawing explanation

Fig. 1 is HEVC reconstruction video " flicker " schematic diagram.

Fig. 2 is the reconstruction video objective quality schematic diagram after Container sequence and Boat sequence adopt the encoding and decoding of R-lambda model.

The target bit rate that Fig. 3 distributes for each CTU during I frame target bit rate deficiency.

Fig. 4 is the HEVC reconstruction video flicker restraint method flow chart based on R-lambda Rate Control model.

Embodiment:

Below in conjunction with embodiment, further illustrate the present invention.

In order to suppress HEVC reconstruction video " flicker " phenomenon, improve reconstruction video subjective visual quality, the present invention proposes a kind of HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model.First the method produces the difference degree of " flicker " phenomenon according to zones of different in I frame, classifying to the CTU of I frame, (move CTU, the complicated CTU of texture, the smooth CTU of texture), then different optimization of rate distribution methods is adopted to dissimilar CTU, utilize R-lambda Rate Control model to calculate the QP value of CTU, the subjective visual quality difference between level and smooth I frame and consecutive frame, while suppression " flicker " phenomenon, improve the overall subjective and objective quality of reconstruction video.

Although a new generation video encoding standard HEVC introduces the new technology that quadtree coding block more flexibly divides, is beneficial to intraframe coding based on the infra-frame prediction in direction and type of prediction etc., but still adopt and H.264/AVC similar coding structure, the code efficiency of I frame is still significantly less than P frame and B frame, for needing the Video Applications periodically inserting I frame, the reconstruction quality of I frame is the key factor affecting the overall subjective and objective quality of video.

The present invention is from the angle of Rate Control, by analyzing the deficiency of R-lambda model in target bit rate distribution adopted in current HEVC identifying code, propose the new target bit rate optimizing distribution method based on CTU type, reach the object of suppression " flicker ".Detailed process of the present invention as shown in Figure 4.

Before introducing embodiment, first the parameter that the present invention uses is described.

MV _x: the component of the motion vector of current C TU (x-axis direction) in the horizontal direction;

MV _y: the component of the motion vector of current C TU (y-axis direction) in the vertical direction;

MV _xPrevious: current C TU is at the component of the motion vector (x-axis direction) in the horizontal direction of former frame corresponding position CTU;

MV _yPrevious: current C TU is at the component of the motion vector (y-axis direction) in the vertical direction of former frame corresponding position CTU;

MV _gx: the component of the global motion vector of current I frame (x-axis direction) in the horizontal direction;

MV _gy: the component of the global motion vector of current I frame (y-axis direction) in the vertical direction;

MAD _curCTU: the complexity of current C TU;

MAD _{collocatedCTU}: current C TU is in the complexity of former frame corresponding position CTU;

MAD _intraFrame: the whole two field picture complexity of current I frame;

MAD _{previousFrame}: the whole two field picture complexity of the former frame of current I frame;

T _i: the target bit rate that i-th I frame distributes;

T _i(j): the target bit rate of a jth CTU in i-th I frame, i and j is positive integer;

T _codedCTU: the actual coding code check sum of encoded CTU in current I frame;

N _left: the number of uncoded CTU in i-th I frame;

QP _i(j): the QP value of a jth CTU in i-th I frame;

ε: judge fault-tolerant value during motion CTU, be set as 20 in the present invention;

A, b: linear forecasting parameter, be set as 1 in the present invention.

Launch to illustrate to the present invention below.

In the CTU assorting process of I frame, first judge whether current C TU is motion CTU.In order to reach this purpose, the present invention adopts the method extracting global motion vector.Because the CTU in I frame all adopts intraframe coding method, the motion vector of each CTU can not be calculated, therefore adopt linear prediction mode, according to the motion vector of current C TU at former frame (P frame or B frame) corresponding position CTU, predict the motion vector of current C TU.Because motion vector is a bivector, computation complexity is higher, and therefore, the motion vector of CTU is decomposed into level (x-axis) and vertical (y-axis) both direction by the present invention, calculates its value respectively.As shown in formula (1) and (2).

MV _x＝a·MV _xPrevious

(1)

MV _y＝a·MV _yPrevious

(2)

The MV of all CTU in statistics I frame _xand MV _yprobability of occurrence, by maximum for the probability of occurrence global motion vector MV as I frame _gxand MV _gy.If the motion vector of a CTU meets condition shown in formula (3), then think that this CTU is motion CTU.

(MV _x＜MV _Gx-ε)∨(MV _x＞MV _Gx+ε)∨(MV _y＜MV _Gy-ε)∨(MV _y＞MV _Gy+ε)(3)

After motion CTU extracts, then according to the complexity of CTU and the whole frame complexity of I frame, judge that CTU is the complicated CTU or the smooth CTU of texture of texture.Because now the complexity of CTU and the whole frame complexity of I frame also do not have Practical Calculation, in order to reduce the computational complexity of the inventive method, allow the inventive method can adapt to real-time video application, the complexity of CTU and the whole frame complexity of I frame are all according to the corresponding complexity information of former frame, obtained by linear prediction, as shown in formula (4) and (5).

MAD _curCTU＝b·MAD _{collocatedCTU}(4)

MAD _intraFrame＝b·MAD _{previousFrame}(5)

MAD _curCTUand MAD _intraFrameafter doping, both are compared, if MAD _curCTU> MAD _intraFrame, then think that current C TU is the complicated CTU of texture, otherwise be the smooth CTU of texture.

Utilizing said method, be divided three classes by the CTU in I frame, is motion CTU, the complicated CTU and the smooth CTU of texture of texture respectively.Then calculate the residue target bit rate of I frame, distribute target bit rate for different CTU optimizes.Why will calculate the residue target bit rate of I frame, be because in video coding process, when I frame residue target bit rate is less than or equal to 0, formula (6) will be adopted to calculate the QP value of CTU.

QP _i(j)＝QP _i(j-1)+1(6)

Along with QP value increases, the reconstruction quality of CTU can decline gradually, the whole body reconstruction quality of I frame is caused to having a strong impact on.The present invention is directed to this problem, at every turn before distributing target bit rate for CTU, all will recalculate the residue target bit rate of I frame.When I frame residue target bit rate is greater than 0, " flicker " phenomenon produced due to motion CTU and the complicated CTU of texture is not obvious, therefore, for this two class CTU distributes less target bit rate; " flicker " phenomenon produced due to the smooth CTU of texture is the most obvious, therefore, for this kind of CTU distributes more target bit rate.The concrete distribution method of target bit rate is as follows, and for motion CTU, target bit rate distributes formula as shown in (7).

$T_i\left(j\right)=\left\{\begin{array}{cc}\frac{{\mathrm{MV}}_{Gx}}{{\mathrm{MV}}_x}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}& {\mathrm{MV}}_x\&GreaterEqual;{\mathrm{MV}}_y\\ \frac{{\mathrm{MV}}_{Gy}}{{\mathrm{MV}}_y}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}& {\mathrm{MV}}_x<{\mathrm{MV}}_y\end{array}\right.---\left(7\right)$

For the complicated CTU and the smooth CTU of texture of texture, target bit rate distributes formula as shown in (8).

$T_i\left(j\right)=\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTU}}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}---\left(8\right)$

When I frame residue target bit rate is less than 0, in order to ensure I frame reconstruction quality, exceeds consumption target bit rate simultaneously, adopting following method to recalculate the target bit rate of all kinds of CTU.For motion CTU and the complicated CTU of texture, ensure its reconstruction quality, target bit rate distributes formula as shown in (9) as far as possible.

$T_i\left(j\right)=\frac{1}{j-1}\&CenterDot;T_{codedCTN}---\left(9\right)$

For the smooth CTU of texture, " flicker " phenomenon as far as possible suppressing it to produce, target bit rate distributes formula as shown in (10).

$T_i\left(j\right)=\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTN}}\&CenterDot;\frac{1}{j-1}\&CenterDot;T_{codedCTU}---\left(10\right)$

In order to avoid the MAD of some CTU _curCTUand MAD _intraFramehave big difference, cause its target bit rate excessive, the present invention is to the T of the smooth CTU of texture _ij () limits, as shown in formula (11).

$T_i\left(j\right)=min\{\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTU}}\&CenterDot;\frac{1}{j-1}\&CenterDot;T_{codedCTU},5\&CenterDot;\frac{1}{j-1}\&CenterDot;T_{codedCTN}\}---\left(11\right)$

After in I frame, the target bit rate distribution of each CTU terminates, R-lambda model is adopted to calculate the QP value of CTU, ensure that the subjective reconstruction quality of I frame can be level and smooth with consecutive frame, suppress " flicker " phenomenon, simultaneously, the target bit rate of I frame consumption is unlikely to again excessive, thus improves the whole body reconstruction quality of video.

Specific embodiments of the invention are only above.According to technological thought provided by the invention, those skilled in the art can think and change, all should fall within the scope of protection of the present invention.

Claims (6)

1., based on a HEVC reconstruction video flicker restraint method for R-lambda Rate Control model, it is characterized in that, the step of the method is as follows:

(1), for intracoded frame, i.e. I frame, predicts movable information and the complexity information of current code tree unit CTU to be encoded, and the whole frame complexity information of current I frame;

(2), according to the movable information of CTU, extract the global motion information of I frame, and judge whether current C TU is motion CTU;

(3), for non-athletic CTU, according to the complexity information of current C TU and the complexity information of I frame, judge that current C TU is the complicated CTU of texture, or the smooth CTU of texture;

(4), according to different CTU types, distribute target bit rate for CTU optimizes, and adopt R-lambda model to calculate the QP value of CTU.

2. the HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model according to claim 1, it is characterized in that, described step (1) specifically comprises the steps:

(11), obtain the CTU of current CTU to be encoded in former frame corresponding position, extract its movable information;

(12), linear prediction method is adopted, the movable information of prediction current C TU;

(13), obtain the whole frame complexity information of the former frame of I frame, and current CTU to be encoded is in the complexity information of former frame corresponding position CTU;

(14), linear prediction method is adopted, the whole frame complexity information of prediction current I frame and the complexity information of current C TU.

3. the HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model according to claim 1, it is characterized in that, described step (2) specifically comprises the steps:

(21), the motion vector of each CTU is decomposed into horizontal motion vector MV _xwith vertical motion vectors MV _y;

(22) MV of each CTU, is added up respectively _xand MV _y, calculate the MV occurring maximum probability _xand MV _y, it can be used as the global motion vector MV of horizontal direction and vertical direction _gxand MV _gy;

(23), by the MV of each CTU _xand MV _ywith MV _gxand MV _gycompare, judge whether current C TU is motion CTU.

4. the HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model according to claim 1, it is characterized in that, described step (4) specifically comprises the steps:

(41) the residue target bit rate of current I frame, is calculated;

(42), when remaining target bit rate and being greater than 0, being motion CTU distribution target bit rate according to the motion vector value of CTU and the motion vector value of I frame, according to the complexity information of CTU and the complexity information of I frame, is that non-athletic CTU distributes target bit rate;

(43), when remaining code check and being less than or equal to 0, calculating the actual coding code check average of current encoded CTU, is that current C TU distributes target bit rate according to this average.

5. the HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model according to claim 4, is characterized in that,

In described step (42), the concrete grammar distributing target bit rate for motion CTU is:

$T_i\left(j\right)=\left\{\begin{array}{cc}\frac{{\mathrm{MV}}_{Gx}}{{\mathrm{MV}}_x}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}& {\mathrm{MV}}_x\&GreaterEqual;{\mathrm{MV}}_y\\ \frac{{\mathrm{MV}}_{Gy}}{{\mathrm{MV}}_y}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}& {\mathrm{MV}}_x<{\mathrm{MV}}_y\end{array}\right.$

The concrete grammar distributing target bit rate for the complicated CTU of texture and the smooth CTU of texture is:

$T_i\left(j\right)=\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTU}}\&CenterDot;\frac{T_i-T_{codedCTU}}{N_{left}}.$

6. the HEVC reconstruction video flicker restraint method based on R-lambda Rate Control model according to claim 4, is characterized in that, in described step (43), the concrete grammar distributing target bit rate for motion CTU and the complicated CTU of texture is:

$T_i\left(j\right)=\frac{1}{j-1}\&CenterDot;T_{codedCTU}$

The concrete grammar distributing target bit rate for the smooth CTU of texture is:

$T_i\left(j\right)=\frac{{\mathrm{MAD}}_{\mathrm{int}raFrame}}{{\mathrm{MAD}}_{curCTU}}\&CenterDot;\frac{1}{j-1}\&CenterDot;T_{codedCTU}.$