CN106791848A - A kind of Two Pass bit rate control methods based on HEVC - Google Patents
A kind of Two Pass bit rate control methods based on HEVC Download PDFInfo
- Publication number
- CN106791848A CN106791848A CN201611183575.7A CN201611183575A CN106791848A CN 106791848 A CN106791848 A CN 106791848A CN 201611183575 A CN201611183575 A CN 201611183575A CN 106791848 A CN106791848 A CN 106791848A
- Authority
- CN
- China
- Prior art keywords
- bit
- lcu
- rate
- frame
- video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/146—Data rate or code amount at the encoder output
- H04N19/147—Data rate or code amount at the encoder output according to rate distortion criteria
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/103—Selection of coding mode or of prediction mode
- H04N19/114—Adapting the group of pictures [GOP] structure, e.g. number of B-frames between two anchor frames
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
- H04N19/124—Quantisation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/189—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the adaptation method, adaptation tool or adaptation type used for the adaptive coding
- H04N19/19—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the adaptation method, adaptation tool or adaptation type used for the adaptive coding using optimisation based on Lagrange multipliers
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention relates to technical field of video coding, more particularly to a kind of Two Pass bit rate control methods based on HEVC.The method is concretely comprised the following steps:Step 1, quick pre encoded video sequence, and collect the statistical information of correspondence coding unit;Step 2, according to the statistical information above collected, will carry out bit distribution, the fitting of rate-distortion model parameter training and the calculating of quantization parameter QP.Present invention has the advantages that:Under the different target bit rate of height, video sequence average peak signal to noise ratio PSNR of the present invention different for complexity, resolution ratio has lifting, averagely improves 0.42dB;In terms of rate control accuracy, code check error of the invention has reduction for different video sequence, averagely reduces 4.81%.
Description
Technical field
The present invention relates to technical field of video coding, more particularly to a kind of Two-Pass Rate Control sides based on HEVC
Method.
Background technology
With the continuous progress of video coding technique, major video develops to high definition and ultra high-definition direction, brings therewith
The surge of the video data volume, in order to reach compression performance higher, JCT-VC joints group is proposed H.265/HEVC video and compiles
Code standard, H.264/AVC the compression performance of standard improves one times than ever.In the practical application of video encoding standard, code
Rate is controlled, i.e., optimal video quality is obtained under set target bit rate, is indispensable key component in Video coding,
It is also the focus of field of video encoding research.
In the practical application of coding and decoding video, it is possible to use bandwidth resources be often limited.Answering per frame video
, all as video content is being continually changing, video complexity is higher, and the number of coded bits of output also can be higher, if not right for miscellaneous degree
Code stream is controlled by, and the video code rate obtained after coding will be continually changing, and the code check being widely varied can not meet actual biography
The requirement of defeated channel and transmission bandwidth, so, Rate Control is essential in the practical application of video encoding standard.
Rate Control is under current bandwidth limitation, target bit rate to be matched by certain bit rate control method and is obtained
A kind of technological means of optimal video quality.Rate-distortion optimization (Rate-Distortion Optimization, RDO) be exactly
Optimal coding parameter is selected under the guidance of rate distortion theory, is taken between the distortion caused after encoding bit rate output and compression
Balance most preferably is obtained, i.e., reduces code check as far as possible on the premise of video quality is ensured, or in identical bit rate output condition
Under lift video quality as far as possible.Rate-distortion optimization theory occupies highly important effect in Rate Control.
" the discrete source coding theory under fidelity criterion " that rate distortion theory is most delivered early in nineteen fifty-nine by Shannon and
Gradually paid close attention to by scholar, mainly in the case where certain distortion is allowed, the limit of information source compression be what this ask
Topic, makes definitions and discusses to rate distortion function R (D) first.The scholar and expert of association area constantly study it after
Afterwards, rate distortion theory gradually improvement and systematization.
The quality of video and the size of final output code stream are all mainly influenceed by quantizing process, to same video using not
Same quantization parameter can obtain different code checks and video quality.So, the main purpose of Rate Control is to obtain one group of quantization
Parameter QP, is used to the follow-up coding flow of control, is finally reached the requirement of target bit rate.Wherein, emphasis is exactly to find code
Mapping relations between rate and quantization parameter, and model is set up, so as to obtain required quantization parameter by target bit rate.
The infra-frame prediction and inter-frame prediction method used in video encoding standard so that the dependence between encoding block is very
By force, the separated consideration that their distortion performance can not be independent.Therefore, although rate distortion function provides quantization parameter set
Relation and distortion, code check between, but it is extremely complex that quantization parameter is directly solved using rate distortion function.Actual
In coding standard application, Rate Control is generally divided into two steps and carries out, as shown in figure 1, wherein, the first step is bit distribution
Target bit rate, is exactly assigned to image sets (GOP), frame (Frame) and volume by (Bit Allocation) by certain rule
On the image as unit such as code block (CTU);Second step is exactly to set up suitable rate-distortion model to calculate quantization parameter (QP), after control
Continuous quantizer so that the code check of final output can match given target bit rate.
(1) bit distribution
Maximum encoding block (Largest Coding Unit, LCU) is H.265/HEVC can be with the base of absolute coding in standard
This unit, H.264/AVC in standard correspondingly be macro block (Micro Block, MB).For HEVC standard, Rate Control
The minimum unit that can be controlled, i.e., can be that its minimum unit for individually distributing quantization parameter be also LCU.The first of Rate Control
Step is Data Rate Distribution, i.e. bit distribution.Bit distribution is progressively segmented, and common practices is as shown in Fig. 2 first by under target bit rate
Total bit number first distributed into GOP image sets by certain rule, then successively by Data Rate Distribution to each two field picture, and then distribute
To LCU grades.Meanwhile, an elementary area has often been encoded in every one-level, it is right just according to the situation for encoding actually used bit number
Subsequent bits distribution is adjusted and updates, i.e., follow-up bit distribution is updated using remaining bits number.
(2) calculating of quantization parameter QP
The second step of Rate Control is that the result for obtaining bit distribution by the first step goes to try to achieve one group of quantization parameter QP, emphasis
It is exactly the foundation to rate distortion relational model between code check and quantization parameter QP.The precision of rate-distortion model is final defeated decision
Whether the code stream for going out can reach the key of target bit rate requirement, have decisive significance to the performance of rate control module.Video
Domain expert and scholar conduct extensive research to rate-distortion model, establish the approximate model of many rate distortion relations, main
Will be by first-order linear model, secondary model, exponential model, logarithmic model, R- λ models etc..
The content of the invention
In order to solve the problems, such as prior art, the present invention provides a kind of HEVC Rate Control sides based on Two-Pass
Method.
The technical scheme is that:A kind of Two-Pass bit rate control methods based on HEVC, the specific step of the method
Suddenly it is:
Step 1, quick pre encoded video sequence, and collect the statistical information of correspondence coding unit;
Step 2, will according to the statistical information above collected, carry out bit distribution, the fitting of rate-distortion model parameter training with
And the calculating of quantization parameter QP.
A kind of Two-Pass bit rate control methods based on HEVC are further comprising the steps of:Would generally be first before Video coding
It is divided into several small GOP image sets, a GOP image sets to generally include an at least two field picture;Generally enter one per two field picture
Step is divided into some code blocks (LCU), so as to further encode.
Described image has three kinds of classifications:I frames, P frames and B frames;Wherein, I frames are intracoded frame, can only refer to this two field picture
Interior pixel is predicted, while can be the key frame in predictive coding as P frames and the reference frame of B frames, the present invention be carried
The bit rate control method for going out will pay close attention to the Rate Control of I frames, i.e. intraframe predictive coding.
The step 1 is comprised the following steps:
Step a:Initialization;For giving video sequence, fixed quantization parameter is configured in the cfg configuration files of coding
QP values, while RateControl parameters are set into 0, close Rate Control switch;
Step b:CU is divided;The given video sequence of input, in function TEncCu::To current video in xCompressCU ()
Each LCU of sequence is recursively divided;
Step c:Judge whether to be divided into 16 × 16;Using the condition whether depth uiDepth variables are less than 3 is divided, sentence
It is disconnected current whether to divide to 16 × 16 levels, such as less than 3, then continue to divide;Otherwise, then prove that current LCU has been divided into 16
× 16 levels, can carry out next step operation;
Step d:Determine whether image boundary;Using border Rule of judgment, image boundary is determined whether, if side
Boundary, it is possible to there is the situation that border LCU can not be divided exactly by 16, then continue to be divided down to 8 × 8 pieces, then carries out follow-up
Prediction, transition coding process;
Step e:Collect the statistical information of the video sequence that coding is produced;It is important that to the bit number shared by each LCU block,
Bit number, video frequency output code check, the collection of video distortion information of whole frame;Wherein, the bit number of the bit number of LCU and whole frame
Information, can use as complexity when subsequent bits are distributed;Code check, distortion information can then constitute (R, D) sample point,
Fitting for subsequent parameter is used;
Step f:Change QP values, the cataloged procedure of repeat step a to step e, until traversal 22,27,32,37 4 kind of QP
Value.
The step 2 includes that bit distribution, rate-distortion model are set up and quantization parameter QP calculates three major parts, specifically
Step is:
Step I:Frame level bit is distributed;In TEncRCSeq::In create () function, according to target bit rate and current sequence
Frame per second, calculate target bit, and target bit is averagely allocated to each frame;
Step II:Modification TEncCu::UpdateLCUDataISlice () function, by the complexity benchmark of LCU by
The SATD of 64 8 × 8 originally and, be revised as fixing when QP is 22 the LCU bit numbers collected in step, that is, define file pointer
Function return value is assigned to what is arranged by raster scan order successively with the bit number information of the reading LCU from file for variable
The complexity variable m_costIntra of LCU;
Step III:Each LCU in traversal present frame, by the cumulative of the complexity variable m_costIntra of all LCU
Be assigned to m_totalCostIntra variables, the variable represents total complexity of present frame;
Step IV:In function TEncRCPic::It is each LCU distributing bit in getLCUInitTargetBits (),
Formula is as follows:
TCurrLCU=(TCurrPic-BitHeader-BitCoded)×CCurrLCU
Wherein, CCurrLCUIt is the complexity of current LCU, B1stCurrLCUIt is the ratio of current LCU being collected into First-Pass
Special number, BNotCodedLCUsIt is the bit number of uncoded LCU in the present frame that is collected into step 1;
Step V:Using video code rate, the distortion information collected in step 1, on the basis of hyperbola R-D models, fitting
Model parameter C, K of current video sequence;
Step VI:According to R-D models and the corresponding relation of R- λ models, model parameter α, β of R- λ models is calculated;
Step VII:R- λ models according to on-line training model parameter, using the bit in step IV being current LCU distribution
Number, calculates the λ value for current LCU, and check whether to meet constraint;
Step VIII:According in step VII obtain λ value and formula QP=4.2005ln λ+13.7122 described in λ with
Relation between QP, calculates the QP values for current LCU quantizing process, and checks whether to meet constraint;
Step IV:The QP values collection obtained in applying step VIII, carries out complete cataloged procedure.
Compared with prior art, present invention has the advantages that:
1) under the different target bit rate of height, compared to the standard bit rate control method in HM16.0, patent of the present invention
The video sequence that the Two-Pass fast method average peak signal to noise ratio PSNR of proposition are different for complexity, resolution ratio is equal
There is lifting, averagely improve 0.42dB, the PSNR liftings for video sequence under target bit rate high are the most obvious, reach as high as
3.14dB;
2) in terms of rate control accuracy, the code check error of patent bit rate control method of the present invention is for different video sequence
There is reduction, averagely reduce 4.81%, equally, for the Rate Control error reduction under target bit rate high significantly, most
Height can reduce by 24.6% code check error.
Brief description of the drawings
Fig. 1 Rate Control basic procedures;
Fig. 2 bits distribute basic procedure;
Fig. 3 different types Image Coding output bit number compares;
Fig. 4 First-Pass algorithm flows;
Fig. 5 LCU statistical information storage orders;
Fig. 6 Second-Pass algorithm flows;
Fig. 7 supports the Two-Pass fast algorithm flows of on-line training.
Specific embodiment
The present invention proposes a kind of HEVC bit rate control methods based on Two-Pass, and its specific method is as follows:
Several small GOP image sets would generally be first divided into before Video coding, if a GOP image sets are generally included
Dry two field picture, these images have three kinds of classifications:I frames (Intra Frame), P frames (Predictive Frame) and B frames (Bi-
directional Interpolate Prediction Frame).Wherein, P frames are forward-predictive-coded frames, before may be referred to
The encoded frame in face carries out inter prediction, eliminates the time redundancy between successive frame;B frames are bi-directional predicted interpolated value coded frames, both
May be referred to the pixel of above encoded frame, it is also possible to encoded with reference to frame below;I frames are intracoded frame, Zhi Nengcan
The pixel examined in this two field picture is predicted, and can simultaneously serve as the reference frame of P frames and B frames, is the key frame in predictive coding,
Occupy highly important position in predictive coding, each GOP image sets is generally using I frames as the first frame.I frames include a width
The all important information of image, can independently be decoded, and the decoding of P frames and B frames will depend on I frames ability above complete
Into.Meanwhile, in whole compressed video stream, I frames can take more bit, in some instances it may even be possible to be several times as much as P frames and B frames.It is such as attached
Fig. 3 show " RaceHorses " sequence (IBBBPBBB structures) under fixed quantisation parameter, and different images type coding is exported
The comparing of bit number.I frames coding output bit number at most, about the 2 of P frames times, 10 times of B frames.In summary as can be seen that
I frames occupy highly important status in whole coded video sequence, and Rate Control is carried out to I frames, can be to whole video sequence
Compression efficiency have vital influence.Therefore, bit rate control method proposed by the invention will pay close attention to I frames, i.e. frame
The Rate Control of intraprediction encoding.
The present invention is broadly divided into two steps:The quick pre encoded video sequence in step 1 (First-Pass), and collect
The statistical informations such as the complexity of correspondence coding unit;In step 2 (Second-Pass), by according to the statistics letter above collected
Breath, carries out bit distribution, the fitting of rate-distortion model parameter training and the calculating of quantization parameter QP.
1. step 1 (First-Pass)
Step 1 (First-Pass) coding is the first step of Two-Pass bit rate control methods proposed by the present invention, mainly
The statistical information important for collecting video sequence.For on-line training model parameter, in the modified Two- that the present invention is carried
, it is necessary to carry out multiple precoding in Pass bit rate control methods, to obtain multigroup real code check and the mistake of current video sequence
Genuine corresponding relation, and then it is fitted the model parameter for adapting to current video sequence.Multiple precoding will cause First-Pass's
The binary encoding time is multiplied, and coding rate has been dragged slowly significantly.Therefore, the present invention proposes a kind of method of quick precoding,
16 × 16 block of pixels will be all divided into per two field picture in step 1 (First-Pass), i.e., only to dividing the CU blocks that depth is 3
Precoding is carried out, statistical information is collected, wherein emphasis is considered border condition in implementation process.The present invention is marked using HM
Under different quantization fine degrees that what quasi- software model was recommended can most reflect the compression effectiveness of video four kinds of fixed QP values (22,
27th, 32,37) encoded, that is, carried out four times quick precodings, obtained 4 couple (R, D) sample point of each frame of current sequence, used
In the fitting of follow-up model parameter;Meanwhile, encoded in the case where quantization parameter QP values take 22, by 16 × 16 in each LCU
The bit information that block of pixels is counted on is added, as the complexity information of the LCU blocks, to carry out follow-up Second-Pass
Coding.The step of using quick precoding 1 (First-Pass) flow is as shown in Figure 4.
In improved Two-Pass fast methods, the realization of First-Pass codings mainly has following steps:
Step a:Initialization.For giving video sequence, fixed quantization parameter is configured in the cfg configuration files of coding
QP values, while RateControl parameters are set into 0, close Rate Control switch;
Step b:CU is divided.The given video sequence of input, in function TEncCu::To current video in xCompressCU ()
Each LCU of sequence is recursively divided;
Step c:Judge whether to be divided into 16 × 16.Using the condition whether depth uiDepth variables are less than 3 is divided, sentence
It is disconnected current whether to divide to 16 × 16 levels, such as less than 3, then continue to divide;Such as it is not less than 3, then proves that current LCU has been divided
To 16 × 16 levels, next step operation can be carried out;
Step d:Determine whether image boundary.Using border Rule of judgment, image boundary is determined whether, if side
Boundary, it is possible to there is the situation that border LCU can not be divided exactly by 16, then continue to be divided down to 8 × 8 pieces, then carries out follow-up
Prediction, conversion Equal-coding pass;
Step e:Collect the statistical information of the video sequence that coding is produced.It is important that to the bit number shared by each LCU block,
Bit number, video frequency output code check, the collection of video distortion information of whole frame.Wherein, the bit number of the bit number of LCU and whole frame
Information, can use as complexity when subsequent bits are distributed;Code check, distortion information can then constitute (R, D) sample point,
Fitting for subsequent parameter is used.
Step f:Change QP values, the cataloged procedure of repeat step a to step e, until traversal 22,27,32,37 4 kind of QP
Value.
Wherein, storage order of the statistical information of each LCU in output file is suitable according to the raster scanning shown in accompanying drawing 5
Sequence, stores successively from left to right, from top to bottom.Meanwhile, the LCU bit numbers collected in First-Pass account for the ratio of whole frame bit number
Rate, the elaborate degree benchmark that will be used as patent of the present invention, for bit distribution in follow-up Second-Pass codings
When use.
Additionally, the distortion information of video is represented using MSE, MSE be rebuild video data and original video sequence data it
Between mean square deviation, shown in its computing formula such as formula (1).Wherein, the original value of pixel is represented, the reconstructed value of pixel is represented, n is represented
Pixel total number.
The code rate information of video represented using bpp, i.e., the bit number shared by average each pixel, its computing formula such as formula
(2) shown in.Wherein, bit represents the total bit number of present frame, PtotalThe total pixel number of present frame is represented, w represents the figure of present frame
Image width degree, h represents the picture altitude of present frame.
Ptotal=w × h (2)
2. step 2 (Second-Pass)
Step 2 (Second-Pass) coding is the second step of the Two-Pass bit rate control methods that patent of the present invention is proposed,
It is the committed step of Two-Pass methods.Step 2 (Second-Pass) coding is main to have bit distribution, rate-distortion model to set up
Three major parts are calculated with quantization parameter QP, the specific method and implementation of each part will be in detail situated between below
Continue:
(1) bit distribution
For intraframe predictive coding, the parameter setting of full I frames is 1 equivalent to each GOP size, wherein only including one
Two field picture, that is, 1 I frame.So, for frame in Rate Control, according to the strategy that layering bit is distributed, GOP grades of ratio
Spy's distribution is equivalent to be frame level bit distribution that the bit number that will be removed under the target bit rate outside header is evenly distributed to often
One I two field picture.
The pixel that I two field pictures can only refer to the encoded CU blocks of this frame is predicted coding, therefore, compare in frame in Rate Control
The emphasis of special allocation step is exactly LCU grades of Bit distribution method.Here, patent of the present invention uses elaborate degree benchmark
It is each LCU distributing bit as weight.I.e. using in First-Pass collect each LCU bit number account for correspondence frame always than
The ratio of special number as the LCU complexity, be its distributing bit under current goal code check.Assuming that the general objective ratio of present frame
Special number is that bit number is shared by header, and the encoded actually used total bit number of block of pixels is, then the ratio that current LCU can get
Special number can be represented with formula (3):
TCurrLCU=(TCurrPic-BitHeader-BitCoded)×CCurrLCU
Wherein, CCurrLCUIt is the complexity of current LCU, B1stCurrLCUIt is the ratio of current LCU being collected into First-Pass
Special number, BNotCodedLCUsIt is the bit number of uncoded LCU in the present frame that is collected into First-Pass.
(2) rate distortion modeling
The target of video encoding standard is to reach definition higher using less bit number all the time, and this target can
To be expressed as Constrained problem.Solving the problems, such as an effective method of the constrained rate-distortion optimization is, bright using glug
Day multiplier optimization (Lagrange Operator Algorithm), constrained optimization problem is converted into unconstrained optimization and is asked
Topic.After introducing Lagrange multiplier, Constrained problem can be converted into the form of formula (4).
I.e.:
If the rate-distortion optimization problem in Video coding being regarded as optimal balance being obtained between code check and distortion,
So, Lagrange multiplier λ is exactly to adjust the key balanced between code check and distortion.From froming the perspective of in physical significance, Lagrange multiplies
Sub- λ is exactly the slope of rate distortion curve.
H.265/HEVC by the R-D models of hyperbolic form in standard classical Rate Control motion K0103, such as formula (6)
It is described, the relation between code check R and Lagrange multiplier λ is deduced, final foundation is rate-distortion model, as described in formula (7),
Wherein, α and β are unified model parameter, and empirical value 3.2003 and -1.367 is taken respectively.The Two-Pass code checks that the present invention is realized
Using the R- λ models as rate-distortion model of the present invention in control method.But unified model parameter is not used, but is utilized
By the repeatedly statistical information such as code check, distortion of video sequence of quick precoding collection in step 1 (First-Pass), for
Each video sequence, on-line training adapts to the model parameter of each video sequence.
D (R)=CRK (6)
In step 2 of the invention (Second-Pass) coding stage, using the code collected in step 1 (First-Pass)
The statistical information of rate R and distortion D, corresponding 4 groups of (R, D) sampling points can be obtained to each video sequence, with formula (6) as mould
Type, the sampling point group to each sequence carries out parameter fitting, can obtain the empirical value of the model parameter C and K of each video sequence.
Further according to R-D models and the relation of R- λ models, as described in formula (7), the pass between parameter C, K and parameter alpha, β can be derived
System, as described in formula (8) and formula (9).
α=- CK (8)
β=K-1 (9)
Then, can be calculated according to formula (8) and formula (9) be adapted to current video sequence R- λ models model parameter α and
β, substitutes into formula (7), can not only be used for the on-line training R- λ models of current video sequence.
(3) quantization parameter QP is calculated
By above-mentioned on-line training process, using the R- λ models for obtaining, present encoding list is partially distributed to according to (1st)
The code check of unit, calculates corresponding λ value, further according to the corresponding relation between λ and quantization parameter QP, as described in formula (10), calculates
QP values are tried to achieve, and for next code process.
QP=4.2005ln λ+13.7122 (10)
Meanwhile, in order to reduce the fluctuation of adjacent encoder cell picture quality, the quantization parameter QP difference controls of adjacent LCU are existed
Within 1, current LCU and its image level QP difference controls are within 2.
In summary introduce and analyze, can by the method for step 2 (Second-Pass) implement step be summarized as with
Under several steps:
Step I:Frame level bit is distributed.In TEncRCSeq::In create () function, according to target bit rate and current sequence
Frame per second, calculate target bit, and target bit is averagely allocated to each frame;
Step II:Modification TEncCu::UpdateLCUDataISlice () function, by the complexity benchmark of LCU by
The SATD of 64 8 × 8 originally and be revised as the LCU bit numbers collected in First-Pass, that is, define file pointer variable with
The bit number information of LCU is read in from file, function return value is assigned to answering by the LCU of raster scan order arrangement successively
Miscellaneous degree variable m_costIntra;
Step III:Each LCU in traversal present frame, by the cumulative of the complexity variable m_costIntra of all LCU
Be assigned to m_totalCostIntra variables, the variable represents total complexity of present frame;
Step IV:In function TEncRCPic::It is every according to the method for formula (3) in getLCUInitTargetBits ()
One LCU distributing bit;
Step V:Using information such as video code rate, the distortions collected in step 1 (First-Pass), with hyperbola R-D moulds
On the basis of type, model parameter C, K of current video sequence is fitted;
Step VI:According to R-D models and the corresponding relation of R- λ models, the model parameter α and β of R- λ models are calculated;
Step VII:R- λ models according to on-line training model parameter, using the bit in step IV being current LCU distribution
Number, calculates the λ value for current LCU, and check whether to meet constraint;
Step VIII:The relation between λ and QP according to described in the λ value and formula (10) obtained in step VII, calculates and uses
In the QP values of current LCU quantizing process, and check whether to meet constraint;
Step IV:The QP values collection obtained in applying step VIII, carries out complete cataloged procedure.
In sum, Two-Pass fast algorithms proposed by the present invention realize flow as shown in fig. 7, implementing step
It is as follows:
Step 1 (First-Pass stages)
Step 1):Input video sequence, to every frame video all only to divide depth be 3, i.e., all be divided into 16 × 16
Code block carries out precoding, if current block is image boundary, continues to be divided into 8 × 8 sizes being encoded again, wherein, precoding
The quantization parameter QP values that process is used are the fixed value being previously set;
Step 2):All 16 × 16 pieces (including the boundary pixel blocks less than 16 × 16) that will be included in each LCU are shared
Bit number is added, and as the bit number shared by current LCU, by Z-shaped scanning sequency, is sequentially output standby in file;
Step 3):The statistical informations such as overall code check, distortion after video sequence coding are collected as follow-up rate-distortion model
The sample point of parameter fitting, and export to standby in file;
Step 4):Change QP values, (present invention takes four kinds of different QP values to repeat above step:22nd, 27,32,37 reality is carried out
Test);
Step 2 (Second-Pass stages)
Step 5):Current goal code check is configured in " encoder_intra_main.cfg " configuration file, by target
Bit number under code check is evenly distributed to each frame;
Step 6):On the basis of the remaining total bit number after bit used by present frame removes encoded LCU, by step 2) in
The complexity weight of the preceding LCU for collecting, the present invention is defined by fixing the statistical information collected when QP is 22, is current LCU distribution
Bit;
Step 7):Using step 3) in the information such as video code rate, distortion collected, on the basis of hyperbola R-D models, intend
Close model parameter C, K of current video sequence;
Step 8):According to R-D models and the corresponding relation of R- λ models, the model parameter α and β of R- λ models are calculated;
Step 9):The R- λ models gone out to the above-mentioned on-line training of present frame application, according to step 6) the middle bit feelings distributed
Condition, calculates the λ value for being applied to current coded unit, and then calculates QP values, and makees a definite limitation to QP values, adjacent to reduce
The video quality fluctuation of coding unit;
Step 10):Using step 9) in the QP values that obtain carry out the coding of entire video sequence, output bit flow.
Experimental data shows, under the different target bit rate of height, compared to the standard rate control algorithm in HM16.0,
Video Two-Pass fast algorithms average peak signal to noise ratio PSNR proposed by the present invention different for complexity, resolution ratio
Sequence has lifting, averagely improves 0.42dB, and the PSNR liftings for video sequence under target bit rate high are the most obvious, highest
Up to 3.14dB;In terms of rate control accuracy, the code check error of rate control algorithm of the present invention is equal for different video sequence
There is reduction, averagely reduce 4.81%, equally, for the Rate Control error reduction under target bit rate high significantly, highest
24.6% code check error can be reduced.
Above-described is only the preferred embodiment of the present invention, it is noted that for a person skilled in the art,
Under the premise of general idea of the present invention is not departed from, some changes and improvements can also be made, these should also be considered as of the invention
Protection domain.
Claims (3)
1. a kind of Two-Pass bit rate control methods based on HEVC, it is characterised in that:The method is concretely comprised the following steps:
Step 1, quick pre encoded video sequence, and collect the statistical information of correspondence coding unit;
Step 2, according to the statistical information above collected, will carry out bit distribution, the fitting of rate-distortion model parameter training and measures
Change the calculating of parameter QP, output bit flow.
2. a kind of Two-Pass bit rate control methods based on HEVC according to claim 1, it is characterised in that:The step
Rapid 1 comprises the following steps:
Step a:Initialization;For giving video sequence, fixed quantization parameter QP is configured in the cfg configuration files of coding
Value, while RateControl parameters are set into 0, closes Rate Control switch;
Step b:CU is divided;The given video sequence of input, in function TEncCu::To current video sequence in xCompressCU ()
Each LCU recursively divided;
Step c:Judge whether to be divided into 16 × 16;Using the condition whether depth uiDepth variables are less than 3 is divided, judge to work as
It is preceding whether to divide to 16 × 16 levels, such as less than 3, then continue to divide;Otherwise, then prove that current LCU has been divided into 16 × 16
Level, can carry out next step operation;
Step d:Determine whether image boundary;Using border Rule of judgment, image boundary is determined whether, if border, have
There may be the situation that border LCU can not be divided exactly by 16, then continue to be divided down to 8 × 8 pieces, then carry out follow-up prediction,
Transition coding process;
Step e:Collect the statistical information of the video sequence that coding is produced;It is important that to the bit number shared by each LCU block, whole
The bit number of frame, video frequency output code check, the collection of video distortion information;Wherein, the bit number letter of the bit number of LCU and whole frame
Breath, can use as complexity when subsequent bits are distributed;Code check, distortion information can then constitute (R, D) sample point, supply
The fitting of subsequent parameter is used;
Step f:Change QP values, the cataloged procedure of repeat step a to step e, until traversal 22,27,32,37 4 kind of QP value.
3. a kind of Two-Pass bit rate control methods based on HEVC according to claim 1 and 2, it is characterised in that:Institute
State step 2 to be set up including bit distribution, rate-distortion model and quantization parameter QP three major parts of calculating, concretely comprise the following steps:
Step I:Frame level bit is distributed;In TEncRCSeq::In create () function, according to target bit rate and the frame of current sequence
Rate, calculates target bit, and target bit is averagely allocated into each frame;
Step II:Modification TEncCu::UpdateLCUDataISlice () function, by the complexity benchmark of LCU by original
64 8 × 8 SATD and, be revised as fixing when QP is 22 the LCU bit numbers collected in step 1, that is, define file pointer and become
Measure to read in the bit number information of LCU from file, function return value is assigned to the LCU arranged by raster scan order successively
Complexity variable m_costIntra;
Step III:Each LCU in traversal present frame, by the cumulative and tax of the complexity variable m_costIntra of all LCU
It is worth and gives m_totalCostIntra variables, the variable represents total complexity of present frame;
Step IV:In function TEncRCPic::It is each LCU distributing bits, formula in getLCUInitTargetBits ()
It is as follows:
TCurrLCU=(TCurrPic-BitHeader-BitCoded)×CCurrLCU
Wherein, CCurrLCUIt is the complexity of current LCU, B1stCurrLCUIt is the bit of current LCU being collected into First-Pass
Number, BNotCodedLCUsIt is the bit number of uncoded LCU in the present frame that is collected into step 1;
Step V:Using video code rate, the distortion information collected in step 1, on the basis of hyperbola R-D models, fitting is current
Model parameter C, K of video sequence;
Step VI:According to R-D models and the corresponding relation of R- λ models, model parameter α, β of R- λ models is calculated;
Step VII:R- λ models according to on-line training model parameter, using the bit number in step IV being current LCU distribution, meter
The λ value for current LCU is calculated, and checks whether to meet constraint;
Step VIII:λ and QP according to described in the λ value and formula QP=4.2005ln λ+13.7122 obtained in step VII it
Between relation, calculate the QP values for current LCU quantizing process, and check whether to meet constraint;
Step IV:The QP values collection obtained in applying step VIII, carries out complete cataloged procedure, output bit flow.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611183575.7A CN106791848B (en) | 2016-12-20 | 2016-12-20 | Two-Pass code rate control method based on HEVC |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611183575.7A CN106791848B (en) | 2016-12-20 | 2016-12-20 | Two-Pass code rate control method based on HEVC |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106791848A true CN106791848A (en) | 2017-05-31 |
CN106791848B CN106791848B (en) | 2020-01-03 |
Family
ID=58895805
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611183575.7A Active CN106791848B (en) | 2016-12-20 | 2016-12-20 | Two-Pass code rate control method based on HEVC |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106791848B (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107277520A (en) * | 2017-07-11 | 2017-10-20 | 中国科学技术大学 | The bit rate control method of infra-frame prediction |
CN107592535A (en) * | 2017-08-18 | 2018-01-16 | 西安邮电大学 | H.265/HEVC image layer bit rate control method |
CN109120934A (en) * | 2018-09-25 | 2019-01-01 | 杭州电子科技大学 | A kind of frame level quantization parameter calculation method suitable for HEVC Video coding |
CN110324615A (en) * | 2018-03-30 | 2019-10-11 | 华为技术有限公司 | A kind of code rate allocation method and device |
CN110351555A (en) * | 2018-04-03 | 2019-10-18 | 朱政 | Multipass based on intensified learning goes through video frequency coding rate distribution and control optimization method |
CN113382242A (en) * | 2021-05-20 | 2021-09-10 | 北京大学深圳研究生院 | Code rate control method, device, equipment and computer readable storage medium |
CN114466189A (en) * | 2021-12-27 | 2022-05-10 | 浙江大华技术股份有限公司 | Code rate control method, electronic device and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060256857A1 (en) * | 2005-05-16 | 2006-11-16 | Douglas Chin | Method and system for rate control in a video encoder |
CN101287123A (en) * | 2008-05-23 | 2008-10-15 | 清华大学 | Code rate controlling method for video coding based on Rho domain |
CN103067720A (en) * | 2013-02-05 | 2013-04-24 | 南京信息工程大学 | Macroblock layer rate control method based on H264 video encoding standard |
CN104104951A (en) * | 2014-07-30 | 2014-10-15 | 北京邮电大学 | Rate control method for screen content coding in HEVC |
CN104159095A (en) * | 2014-02-19 | 2014-11-19 | 上海大学 | Code rate control method for multi-view texture video and depth map coding |
-
2016
- 2016-12-20 CN CN201611183575.7A patent/CN106791848B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060256857A1 (en) * | 2005-05-16 | 2006-11-16 | Douglas Chin | Method and system for rate control in a video encoder |
CN101287123A (en) * | 2008-05-23 | 2008-10-15 | 清华大学 | Code rate controlling method for video coding based on Rho domain |
CN103067720A (en) * | 2013-02-05 | 2013-04-24 | 南京信息工程大学 | Macroblock layer rate control method based on H264 video encoding standard |
CN104159095A (en) * | 2014-02-19 | 2014-11-19 | 上海大学 | Code rate control method for multi-view texture video and depth map coding |
CN104104951A (en) * | 2014-07-30 | 2014-10-15 | 北京邮电大学 | Rate control method for screen content coding in HEVC |
Non-Patent Citations (1)
Title |
---|
YEO-JIN YOON ET AL.: "A New Rate Control Method for Hierarchical Video Coding in HEVC", 《IEEE INTERNATIONAL SYMPOSIUM ON》 * |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107277520A (en) * | 2017-07-11 | 2017-10-20 | 中国科学技术大学 | The bit rate control method of infra-frame prediction |
CN107277520B (en) * | 2017-07-11 | 2020-05-12 | 中国科学技术大学 | Code rate control method for intra-frame prediction |
CN107592535A (en) * | 2017-08-18 | 2018-01-16 | 西安邮电大学 | H.265/HEVC image layer bit rate control method |
CN107592535B (en) * | 2017-08-18 | 2020-09-01 | 西安邮电大学 | H.265/HEVC image layer code rate control method |
CN110324615A (en) * | 2018-03-30 | 2019-10-11 | 华为技术有限公司 | A kind of code rate allocation method and device |
CN110351555A (en) * | 2018-04-03 | 2019-10-18 | 朱政 | Multipass based on intensified learning goes through video frequency coding rate distribution and control optimization method |
CN109120934A (en) * | 2018-09-25 | 2019-01-01 | 杭州电子科技大学 | A kind of frame level quantization parameter calculation method suitable for HEVC Video coding |
CN109120934B (en) * | 2018-09-25 | 2020-06-23 | 杭州电子科技大学 | Frame-level quantization parameter calculation method suitable for HEVC video coding |
CN113382242A (en) * | 2021-05-20 | 2021-09-10 | 北京大学深圳研究生院 | Code rate control method, device, equipment and computer readable storage medium |
CN114466189A (en) * | 2021-12-27 | 2022-05-10 | 浙江大华技术股份有限公司 | Code rate control method, electronic device and storage medium |
CN114466189B (en) * | 2021-12-27 | 2024-05-28 | 浙江大华技术股份有限公司 | Code rate control method, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN106791848B (en) | 2020-01-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106791848A (en) | A kind of Two Pass bit rate control methods based on HEVC | |
CN101395923B (en) | Method and apparatus for determining in picture signal encoding the bit allocation for groups of pixel blocks in a picture | |
DE69836052T2 (en) | Method and apparatus for image processing, and data recording medium | |
CN100536573C (en) | Inframe prediction method used for video frequency coding | |
CN104796705B (en) | A kind of HEVC Video codings rate-distortion optimization and rate control algorithm based on SSIM | |
KR100304103B1 (en) | Method for finding re-quantization step sizes resulting in abrupt bit-rate reduction and rate control method using it | |
US6912255B2 (en) | Bit rate control through selective modification of DCT coefficients | |
KR100307618B1 (en) | Device and method for encoding image | |
CN103051901B (en) | Video data coding device and method for coding video data | |
CN104320657B (en) | The predicting mode selecting method of HEVC lossless video encodings and corresponding coding method | |
CN107277520B (en) | Code rate control method for intra-frame prediction | |
CN101653003A (en) | Be used for the quantification of hybrid video coding | |
CN104185024B (en) | A kind of HEVC quantization parameter optimization methods based on total bitrate and entropy model | |
CN106231320A (en) | A kind of unicode rate control method supporting multi-host parallel to encode and system | |
CN103152565A (en) | Image decoding method and image coding method | |
CN104333761B (en) | A kind of HEVC elementary cells level bit-rate distribution method | |
CN107222748A (en) | The treating method and apparatus of view data code check | |
CN106657998A (en) | KVM video coding quantization parameter range control method | |
CN102754440A (en) | Image encoding method, image encoding device and imaging system | |
KR19980074651A (en) | Rate control device for MPEG video signal using fuzzy control | |
JP3777002B2 (en) | Video signal encoding method and video signal encoding apparatus | |
CN105430395B (en) | A kind of HEVC CTU level bit-rate control methods based on optimum bit distribution | |
CN100448295C (en) | Integration code rate control method of low complexity | |
JPH10336654A (en) | Image encoder | |
CN100505874C (en) | Video-frequency encoding-rate controlling method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
CB02 | Change of applicant information | ||
CB02 | Change of applicant information |
Address after: 450007 Zhengzhou Zhongyuan Road, Henan, No. 212 Applicant after: China Electric Power Construction Group Henan electric survey and Design Institute Co., Ltd. Address before: 450007 Zhengzhou Zhongyuan Road, Henan, No. 212 Applicant before: Hennan Electric Power Survey & Design Institute |
|
GR01 | Patent grant | ||
GR01 | Patent grant |