CN105554503A - HEVC encoding unit level code rate control method - Google Patents

Abstract

The invention provides an HEVC encoding unit level code rate control method. An encoding unit SubCU further divided by LCU is used for carrying out level code rate control initialization, bit number allocation, encoding parameter estimation and other processes of P frame and B frame encoding units. At first, former LCU encoding information of the same level frame and at the same position is used for predicating whether the current frame LCU is divided. The LCU predicted to be divided, and an R-D model of the SubCU is used for estimating the complexity and the weight in the encoding unit level code rate control initialization part. In a bit number allocation phase, the state of an encoder and the complexity of the current LCU are respectively considered to calculate a target bit number of the current LCU. In an encoding parameter estimation phase, the R-D model of the SubCU is used for estimating the encoding parameter of the current LCU.

Description

A kind of HEVC coding unit level bit-rate control method

Technical field

The present invention relates to high-performance video coding (HighEfficiencyVideoCoding, HEVC) field, particularly wherein coding unit level bit-rate control method.

Background technology

Rate control algorithm, as a key technology in Video coding application, has important value for practical application.In actual applications, rate control algorithm can according to target bit rate to Video coding, and after making Video coding, actual bit rate is close to target bit rate, improves the quality after coding as much as possible simultaneously.Have employed new coding structure such as coding unit (CodingUnit, CU) and such as the self adaptation sampling point compensation etc. of new technology in HEVC and improve code efficiency.Accordingly, new coding unit level bit-rate control algolithm in HEVC, should be taked, to be adapted to the new coding structure of HEVC and technology, improve the quality after Video coding as far as possible.

The size of the CU that the testing software (HEVCTestModel, HM) of HEVC adopts is 64 × 64,32 × 32,16 × 16 and 8 × 8, and the degree of depth of corresponding CU is 0,1,2,3 respectively.Wherein the degree of depth be 0 CU be designated as LCU (LargestCodingUnit), the degree of depth be 1 CU be designated as SubCU (SubCodingUnit).The encoder of HM adopts the mode of recurrence to divide LCU, and the CU for every one deck will carry out RDO and determine whether being divided into lower one deck.Usually, in order to reduce bit number shared by coding parameter, the target of coding unit level bit-rate control algolithm is general only for LCU, also can be called LCU level bit-rate control algolithm.

Rate control algorithm is divided into three ranks by HEVC existing standard: picture group level, frame level, coding unit level, and have employed a series of coding unit level bit-rate control method: for the LCU in P, B frame, use the relation of R-λ model description bit number and coding parameter.The principle of LCU being carried out to bits allocation is expect that the coding parameter λ of each LCU in frame of video is equal.On this basis, with present frame target bit for constraints, use the R-λ model of each LCU, the bit number that each LCU should get can be solved.Detailed realization can see document 1 (see LiB, LiH, LiL, etal.RatecontrolbyR-lambdamodelforHEVC [C] //JCTVC-K0103, JCTVCofISO/IECandITU-T, 11thmeetingShanghai, China.2012) and document 2 (see LJCTVC-M0036, " AdaptivebitallocationforR-lambdamodelratecontrolinHM, " Incheon, KR, 18 – 26Apr.2013).Compared with standard comparatively early, the method can improve coding quality, but does not consider partition structure and the texture features of LCU, still has the space of improvement.Document 3 is (see JCTVC-M0257, " IntraFrameRateControlBasedonSATD; " Incheon, KR, 18 – 26Apr.2013) propose in I frame, use the texture features of rear absolute length chang (SumofAbsoluteTransformedDifferences, SATD) the description encoding unit of conversion.And use based on the R-λ model description bit number of SATD and the relation of coding parameter.Its SATD value is used to carry out bits allocation as weight to LCU in I frame.The SubCU level bit-rate that the program only considered I frame distributes, and does not design corresponding scheme to P frame and B frame.

Summary of the invention

The object of the present invention is to provide a kind of HEVC coding unit level bit-rate control method.

The present invention need protect technical scheme be:

A kind of HEVC coding unit level bit-rate control method, summarizes and says, at coding unit level bit-rate control algolithm initial phase, in prediction present frame, whether each LCU divides.Whether the LCU according to the same position for previous frame at the same level of prediction divides.If the LCU actual division of the same position of previous frame at the same level, namely predict that this LCU also can divide.Otherwise, as the same.

A kind of HEVC coding unit level bit-rate control method, is characterized in that, specifically comprise the steps:

1) encode before present frame, carry out frame level bit-rate control algorithm, target bit, QP and λ of prediction present frame.Enter coding unit level bit-rate control algolithm initialization section.

2) predict in present frame, whether each LCU divides.If prediction does not divide, calculate weight according to formula (1); Otherwise, calculate weight according to formula (2) (3).

3) InitBits of each LCU is calculated according to formula (4) _currLCU.

4) encoding before current LCU, is current LCU distribution target bit according to formula (5) (6).

5) if prediction does not divide, according to formula (7) calculation code parameter lambda _currLCU; Otherwise, according to formula (8) (9) (10) calculation code parameter lambda _currLCU.

6) QP of current LCU is obtained according to formula (11) _currLCU.

7) according to λ _currLCUand QP _currLCUmodel selection Equal-coding pass is carried out to current LCU.After current LCU end-of-encode, whether be divided into SubCU layer according under actual coding outcome record.If be divided into SubCU layer, according to the actual coding parameter of each SubCU and the R-λ model of result renewal SubCU.Upgrade the R-λ model of LCU.Enter next LCU.

Owing to adopting such scheme, the invention has the beneficial effects as follows:

1. the present invention is according to the partition structure of LCU, relation between the distortion of the LCU using the R-λ model description of SubCU to divide and coding parameter, and propose the method calculating LCU complexity, calculation code parameter according to the R-λ model of SubCU, improve the objective quality after coding.

2. the present invention considers the complexity of encoder current state and LCU, devises the method into current LCU allocation bit number, can be LCU allocation bit number more accurately, improves the objective quality after coding.

Embodiment

Explain below by way of some embodiments, set forth and support the technical scheme that will protect required for the present invention.

Embodiment 1

Theoretical foundation and checking

First, in existing HEVC standard, for P frame and B frame, each LCU sets up, uses and upgrades its R-λ model according to position.But in the ordinary course of things, the content change of LCU inside is very large, and texture features is different, and the bit number that under same coding parameter, different sub-CU obtains has very big difference, so final HEVC encodes, LCU can be divided into multiple little CU by it, carries out the process such as motion compensation, quantization encoding respectively.Now only with a R-λ model, its encoding characteristics is described to single LCU, the feature that LCU inner vein different from those is very large can not be characterized, the partition structure to LCU in actual coding process can not be characterized.But LCU can not form again the statistical property of similar frame level, so describe its encoding characteristics to single LCU with a R-λ model not meet actual conditions.Secondly, for the method for current LCU allocation bit number is desirably in the error making up in frame of video within several LCU and predict bit number and actual bit number in existing standard, easily cause its target bit excessive or too small.

Devise the coding unit level bit-rate control algolithm based on SubCU herein for this reason.

At coding unit level bit-rate control algolithm initial phase, in prediction present frame, whether each LCU divides.Whether the LCU according to the same position for previous frame at the same level of prediction divides.If the LCU actual division of the same position of previous frame at the same level, namely predict that this LCU also can divide.Otherwise, as the same.Table 1 for the method regular coding QP in BQMall sequence be accuracy rate in 32 situations.Test platform is HEVC reference software HM14.0, and test configurations is RandomAccessMain.

Table 1.LCU divides Forecasting Methodology accuracy rate

Frame level	Accuracy rate
		Framelevel＝1	89.16％
Framelevel＝2	85.58％
		Framelevel＝3	87.63％
Framelevel＝4	89.40％
		Mean value	88.53％

As seen in table 1, this LCU divides prediction side and has higher accuracy.Because the method is simply effective, the present invention uses the method to divide Forecasting Methodology as LCU.

At coding unit level bit-rate control algolithm initial phase, for predicting the LCU do not divided, the R-λ model of LCU is used to estimate that its complexity is as weights W eight _currLCU:

${\mathrm{Weigth}}_{currLCU}=\frac{1}{{\mathrm{\α}}_{currLCU}}\×{{\mathrm{\λ}}_{currPic}}^{\frac{1}{{\mathrm{\β}}_{currLCU}}}\×{\mathrm{numberOfPixels}}_{currLCU}---\left(1\right)$

Wherein λ _currPicfor the coding parameter of the present frame that frame level bit-rate control algorithm provides.NumberOfPixels _currLCUfor the number of pixels of current LCU.α _currLCU, β _currLCUfor the parameter of the R-λ model of current LCU.For the LCU divided, the R-λ model of its SubCU is used to estimate that its complexity is as weights W eight _currLCU:

${\mathrm{Weight}}_{currLCU}={\mathrm{\Σ}}_{i=0}^{numberOfSubCU}{\mathrm{InitBits}}_{{\mathrm{SubCU}}_i}---\left(2\right)$

${\mathrm{InitBits}}_{{\mathrm{SubCU}}_i}=\frac{1}{{\mathrm{\α}}_{{\mathrm{SubCU}}_i}}\×{{\mathrm{\λ}}_{currPic}}^{\frac{1}{{\mathrm{\β}}_{{\mathrm{SubCU}}_i}}}\×{\mathrm{numberOfPixels}}_{{\mathrm{SubCU}}_i}---\left(3\right)$

Wherein, for the bits of original number of each SubCU in LCU, representing in λ value is λ _currPicin situation, the bit number needed for each SubCU. it is the parameter of the R-λ model of i-th SubCU. it is the number of pixels of i-th SubCU.NumberOfSubCU is that in current LCU, number of pixels is not the number of the SubCU of 0.

After the weight calculation of all LCU is complete, according to the bits of original number InitBits of the target bit of present frame with the weight calculation LCU of each LCU _currLCU:

${\mathrm{InitBits}}_{currLCU}=\frac{T_{currPic}}{{\mathrm{\Σ}}_{i=0}^{numberOfLCU}{\mathrm{Weight}}_i}\×{\mathrm{Weight}}_{currLCU}---\left(4\right)$

After calculating completes, start the actual coding work of present frame.Before the current LCU of coding, the coding situation weighing present encoding result and prediction is also needed to be that current LCU distributes a suitable target bit TargetBits _currLCU, method is as follows:

TargetBits _currLCU＝a×InitBits _currLCU+(1-a)×LeftBits _currLCU(5)

${\mathrm{LeftBits}}_{currLCU}=\frac{{\mathrm{BitsLeft}}_{currPic}}{{\mathrm{\Σ}}_{i=currLCU}^{numberOfLCU}{\mathrm{Weight}}_i}\×{\mathrm{Weight}}_{currLCU}---\left(6\right)$

Wherein a is weighted value.In order to determine InitBits _currLCUand LeftBits _currLCUweight, invention has been following experiment: respectively different value is got to a, the coding result that test is actual.Cycle tests have chosen RaceHorses, BQSquare, BQMall, ParkScene.Test benchmark is the original rate control algorithm of HEVC reference software HM14.0, and test configurations is RandomAccessMain.Testing algorithm is method of the present invention, and table intermediate value is the mean value of all cycle tests BD-rate values.

Table 2.a value and encoding efficiency relation

Sequence a

0

0.2

0.4

0.5

0.6

0.8

1

BQSquare

-0.68％

-0.68％

-0.52％

0.13％

-0.17％

-0.17％

-1.23％

RaceHorses

-0.48％

-0.52％

-0.73％

-0.69％

-0.63％

-0.81％

-1.07％

BQMall

-1.09％

-1.10％

-1.23％

-0.96％

-1.33％

-1.34％

-0.77％

ParkScene

-0.62％

-1.16％

-0.93％

-0.75％

-0.87％

-0.66％

-0.29％

Mean value

-0.72％

-0.87％

-0.85％

-0.57％

-0.68％

-0.75％

-0.84％

Experimentally result a gets 0.2, to obtain average encoding efficiency preferably.

In coding parameter estimation stages, according to the coding parameter that target bit calculating LCU should use.For predicting the LCU do not divided, the R-λ model of LCU is used to estimate its coding parameter:

${\mathrm{\λ}}_{currLCU}={\mathrm{\α}}_{currLCU}\×\left(\frac{{\mathrm{TargetBits}}_{currLCU}}{{\mathrm{numberOfPixels}}_{currLCU}}\right)\×{\mathrm{\β}}_{currLCU}---\left(7\right)$

For the LCU that prediction divides, the R-λ model of its SubCU is used to estimate its coding parameter:

${\mathrm{TargetBits}}_{{\mathrm{SubCU}}_i}=\frac{{\mathrm{TargetBits}}_{currLCU}}{{\mathrm{InitBits}}_{currLCU}}\×{\mathrm{InitBits}}_{{\mathrm{SubCU}}_i}---\left(8\right)$

${\mathrm{\λ}}_{{\mathrm{SubCU}}_i}={\mathrm{\α}}_{{\mathrm{SubCU}}_i}\×\left(\frac{{\mathrm{TargetBits}}_{{\mathrm{SubCU}}_i}}{{\mathrm{numberOfPixels}}_{{\mathrm{SubCU}}_i}}\right)\×{\mathrm{\β}}_{{\mathrm{SubCU}}_i}---\left(9\right)$

${\mathrm{\λ}}_{currLCU}={\mathrm{\Π}}_{i=0}^{numberOfSubcCU}{{\mathrm{\λ}}_{{\mathrm{SubCU}}_i}}^{\frac{1}{numberOfSubcCU}}---\left(10\right)$

Final coding parameter QP calculates according to following formula:

QP＝4.2005×lnλ+13.7122(11)

In order to ensure the level and smooth of encoding efficiency, λ _currLCU, QP _currLCUits span can be limited according to the state of the encoded part of encoder.Last according to λ _currLCU, QP _currLCUcurrent LCU is carried out to the codings such as model selection.After current LCU end-of-encode, need to upgrade R-D model according to actual coding result, and record current LCU and whether be divided into SubCU layer.If current LCU is divided into SubCU layer really, according to the actual coding parameter of each SubCU and the R-λ model of result renewal SubCU.No matter whether divide, the R-λ model of current LCU all should be upgraded.Then the coding of next LCU is carried out.

Based on above analysis, propose HEVC coding unit level bit-rate control method herein, method text description is as follows:

1) encode before present frame, carry out frame level bit-rate control algorithm, target bit, QP and λ of prediction present frame.Enter coding unit level bit-rate control algolithm initialization section.

2) predict in present frame, whether each LCU divides.If prediction does not divide, calculate weight according to formula (1); Otherwise, calculate weight according to formula (2) (3).

3) InitBits of each LCU is calculated according to formula (4) _currLCU.

4) encoding before current LCU, is current LCU distribution target bit according to formula (5) (6).

5) if prediction does not divide, according to formula (7) calculation code parameter lambda _currLCU; Otherwise, according to formula (8) (9) (10) calculation code parameter lambda _currLCU.

6) QP of current LCU is obtained according to formula (11) _currLCU.

7) according to λ _currLCUand QP _currLCUmodel selection Equal-coding pass is carried out to current LCU.After current LCU end-of-encode, whether be divided into SubCU layer according under actual coding outcome record.If be divided into SubCU layer, according to the actual coding parameter of each SubCU and the R-λ model of result renewal SubCU.Upgrade the R-λ model of LCU.Enter next LCU.

Table 3 is the contrast of the coding result (Pro) after the coding unit level bit-rate control algolithm that open existing coding unit level bit-rate control algolithm in HEVC reference software HM14.0 after, coding result (Orig) and unlatching the present invention propose, wherein BD_rate _yrepresent the coding result of Y-component, BD_rate _urepresent the coding result of U component, BD_rate _vrepresent the coding result of V component, BD_rate is the comprehensive of three:

BD_rate＝(4×BD_rate _Y+BD_rate _U+BD_rate _V)/6(12)

Wherein, the reference data that all BD_rate calculate is that HM14.0 only opens existing frame level bit-rate control, and does not open the coding result of any coding unit level bit-rate control.Can see from table, compare with the existing frame level bit-rate control coding result of HEVC, the coding quality of HEVC existing coding unit level bit-rate control algolithm on average has the loss of 0.07%BD-rate, and institute of the present invention extracting method on average has the coding quality of 1.12%BD-rate to promote.

Table 3. the inventive method (Pro) and the experimental result of the existing LCU code rate allocation method (Orig) of HEVC under RandomAccessMain configuration condition contrast

Embodiment 2

Below in conjunction with attached reproduction technical solution of the present invention embodiment illustrated in fig. 1.

As shown in Figure 1, an embodiment of the present invention comprises the following steps:

Step 1: initialization before coding.For each SubCU and LCU sets up R-λ model, in P, B frame, R-λ model parameter α, β initial value are set to 3.2003 ,-1.367 respectively.

Step 2: before coding present frame, carry out frame level bit-rate control algorithm, target bit, QP and λ of prediction present frame.Enter coding unit level bit-rate control algolithm initialization section.

Step 3: in prediction present frame, whether each LCU divides.If prediction does not divide, calculate weight according to formula (1); Otherwise, calculate weight according to formula (2) (3).

Step 4: the InitBits calculating each LCU according to formula (4) _currLCU.

Step 5: start present frame of encoding.

Step 6: before the current LCU that encodes, according to formula (5) (6) for current LCU distributes target bit.

Step 7: if prediction does not divide, according to formula (7) calculation code parameter lambda _currLCU; Otherwise, according to formula (8) (9) (10) calculation code parameter lambda _currLCU.

Step 8: the QP obtaining current LCU according to formula (11) _currLCU.

Step 9: according to λ _currLCUand QP _currLCUmodel selection Equal-coding pass is carried out to current LCU.After current LCU end-of-encode, whether be divided into SubCU layer according under actual coding outcome record.If be divided into SubCU layer, according to the actual coding parameter of each SubCU and the R-λ model of result renewal SubCU.Upgrade the R-λ model of LCU.Enter next LCU.

Claims (1)

1. a HEVC coding unit level bit-rate control method, is characterized in that, specifically comprise the steps:

1) encode before present frame, carry out frame level bit-rate control algorithm, target bit, QP and λ of prediction present frame; Enter coding unit level bit-rate control algolithm initialization section;

2) predict in present frame, whether each LCU divides; If prediction does not divide, calculate weight according to formula (1); Otherwise, calculate weight according to formula (2) (3);

3) InitBits of each LCU is calculated according to formula (4) _currLCU;

4) encoding before current LCU, is current LCU distribution target bit according to formula (5) (6);

5) if prediction does not divide, according to formula (7) calculation code parameter lambda _currLCU; Otherwise, according to formula (8) (9) (10) calculation code parameter lambda _currLCU;

6) QP of current LCU is obtained according to formula (11) _currLCU;

7) according to λ _currLCUand QP _currLCUmodel selection Equal-coding pass is carried out to current LCU; After current LCU end-of-encode, whether be divided into SubCU layer according under actual coding outcome record; If be divided into SubCU layer, according to the actual coding parameter of each SubCU and the R-λ model of result renewal SubCU; Upgrade the R-λ model of LCU; Enter next LCU.