CN107197296A - A kind of HEVC parallel encoding method and systems based on COStream - Google Patents
A kind of HEVC parallel encoding method and systems based on COStream Download PDFInfo
- Publication number
- CN107197296A CN107197296A CN201710477885.8A CN201710477885A CN107197296A CN 107197296 A CN107197296 A CN 107197296A CN 201710477885 A CN201710477885 A CN 201710477885A CN 107197296 A CN107197296 A CN 107197296A
- Authority
- CN
- China
- Prior art keywords
- image sets
- ctu
- parallel
- entropy code
- costream
- 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
- H04N19/42—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation
- H04N19/436—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation using parallelised computational arrangements
-
- 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
- H04N19/13—Adaptive entropy coding, e.g. adaptive variable length coding [AVLC] or context adaptive binary arithmetic coding [CABAC]
-
- 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
- H04N19/70—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by syntax aspects related to video coding, e.g. related to compression standards
-
- 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
- H04N19/90—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using coding techniques not provided for in groups H04N19/10-H04N19/85, e.g. fractals
- H04N19/91—Entropy coding, e.g. variable length coding [VLC] or arithmetic coding
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Computing Systems (AREA)
- Theoretical Computer Science (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a kind of HEVC parallel encoding method and systems based on COStream, the realization of wherein method includes：Video sequence is divided into some image sets, image sets parallel encoding is carried out by the function for calling C++ to encapsulate in COStream；Each two field picture in image sets is divided into several CTU according to HEVC standard, image parallel encoding is carried out using COStream splitjoin structures；Each CTU includes several syntactic elements, one entropy code function of each syntactic element correspondence, each syntactic element calls the entropy code function of each syntactic element using COStream pipeline, obtains entropy code streamline, and parallel entropy code is carried out to syntactic element using entropy code streamline；Video after being compressed.The present invention solves the demand of magnanimity multimedia video data fast coding under internet environment, reduce the compression time of video data, provided convenience for video service provider quick response client request, while reducing the time that user asks Video service, improve efficiency.
Description
Technical field
It is parallel more particularly, to a kind of HEVC based on COStream the invention belongs to Computer Applied Technology field
Coding method and system.
Background technology
With the popularization and the extensive use of handheld device of internet, the data on internet show volatile increasing
Long, the wherein multi-medium data such as video and image account for significant proportion.The multi-medium data of these magnanimity is passed to storage and network
Defeated to be proposed very high requirement, simple meets growing multimedia by increase storage device with the network bandwidth is improved
Data are unpractical, it is necessary to solve this problem by method for distinguishing.
Traditional programming language such as C++ and Java etc. are proposed eighties of last century the 80s and 90s using single instruction stream as spy
The programming language levied, they can not be applied to the multicore architecture of current generally existing very well.In order to be provided using the calculating of multinuclear
Source, the parallel framework such as pthread, MPI, OpenMP is suggested, but their learning cost is high, user interface is unfriendly, right
The requirement of programmer is very high.
The newest coding standard in coding and decoding video field is HEVC, its coding framework with it is H.264 similar, still using base
In the hybrid encoding frame of block of pixels.In order to tackle under current internet environment the spy that the video data volume is big, video resolution is high
Point, HEVC proposes a series of new coding techniques.Although in the domestic and foreign scholars in terms of the optimization processing of HEVC Video codings
Substantial amounts of research has been carried out, but in the case where current internet multimedia data explosion increases, HEVC Video codings
Still suffer from following Railway Project：The demand of magnanimity multimedia video data fast coding under internet environment is not solved, depending on
The compression time of frequency evidence is longer, and the time of user's request Video service is longer, less efficient.
The content of the invention
For the disadvantages described above or Improvement requirement of prior art, the invention provides a kind of HEVC based on COStream simultaneously
Row coding method and system, thus solve prior art and do not solve magnanimity multimedia video data quick braiding under internet environment
The demand of code, the compression time of video data is longer, and the time of user's request Video service is longer, and less efficient technology is asked
Topic.
To achieve the above object, according to one aspect of the present invention, there is provided a kind of HEVC based on COStream is parallel
Coding method, including：
(1) image sets parallel encoding：Video sequence is divided into some image sets, the information of each image sets is obtained, will
The information of each image sets inputs corresponding image sets encoder, is carried out by the function for calling C++ to encapsulate in COStream
Parallel encoding；
(2) image parallel encoding：Each two field picture in image sets is divided into several CTU according to HEVC standard, collected
The information of CTU rows, a wavefront parallel encoder is inputted by the information of every row CTU rows, is tied using COStream splitjoin
Structure carries out parallel encoding；
(3) the parallel entropy code of syntactic element：Each CTU includes several syntactic elements, each syntactic element correspondence one
Entropy code function, each syntactic element calls the entropy code function of each syntactic element using COStream pipeline, obtains
To entropy code streamline, parallel entropy code is carried out to syntactic element using entropy code streamline；
(4) compression of video sequence：Image sets parallel encoding, image parallel encoding and syntactic element are carried out simultaneously to video sequence
After row entropy code, the video after being compressed.
Further, step (1) includes：
Video sequence is divided into some image sets by (1-1), obtains the information of each image sets, the information of described image group
The size of each two field picture, the start frame of image sets and knot in the quantity of image, image sets in ID, image sets including image sets
Beam frame；
The information of each image sets is inputted corresponding image sets encoder by (1-2), when image sets are open image sets
When, corresponding image sets encoder periodically inserts every IDR frame when coding；
(1-3) carries out parallel encoding in image sets encoder by the function for calling C++ to encapsulate in COStream.
Further, step (2) includes：
Each two field picture in image sets is divided into several CTU by (2-1) according to HEVC standard, collects the information of CTU rows,
CTU number in index of the information of the CTU rows including CTU rows, the line number of CTU rows and CTU rows；
The information of every row CTU rows is inputted a wavefront parallel encoder by (2-2), uses COStream splitjoin
Structure carries out parallel encoding；In addition to the first row CTU rows, remaining CTU row is intended to progress and waited until receive one
The notice of CTU rows, can just be continued executing with；Current wavefront coded device often encodes one CTU of completion and just sent to next CTU rows
Notify.
Further, step (3) includes：
(3-1) each CTU includes several syntactic elements, and syntactic element is converted into after binary bits string carrying out arithmetic volume
Code, one context probability modeling of each syntactic element correspondence, entropy code function uses arithmetic coding and context probability modeling；
(3-2) each syntactic element calls the entropy code function of each syntactic element using COStream pipeline,
Entropy code streamline is obtained, parallel entropy code is carried out to syntactic element using entropy code streamline.
It is another aspect of this invention to provide that there is provided a kind of HEVC parallel encoder systems based on COStream, including：
Image sets parallel encoding module, for video sequence to be divided into some image sets, obtains the letter of each image sets
Breath, corresponding image sets encoder is inputted by the information of each image sets, passes through the function for calling C++ to encapsulate in COStream
Carry out parallel encoding；
Image parallel encoding module, for each two field picture in image sets to be divided into several CTU according to HEVC standard,
The information of CTU rows is collected, the information of every row CTU rows is inputted into a wavefront parallel encoder, uses COStream's
Splitjoin structures carry out parallel encoding；
The parallel entropy code module of syntactic element, several syntactic elements, each syntactic element correspondence are included for each CTU
One entropy code function, each syntactic element calls the entropy code letter of each syntactic element using COStream pipeline
Number, obtains entropy code streamline, and parallel entropy code is carried out to syntactic element using entropy code streamline；
Compression of video sequence module, for carrying out image sets parallel encoding, image parallel encoding and grammer to video sequence
After elements in parallel entropy code, the video after being compressed.
Further, image sets parallel encoding module includes：
First submodule, for video sequence to be divided into some image sets, obtains the information of each image sets, the figure
The size of each two field picture in the quantity of image, image sets in ID, image sets as the information including image sets of group, image sets
Start frame and end frame；
Second submodule, for the information of each image sets to be inputted into corresponding image sets encoder, when image sets are to open
When putting formula image sets, corresponding image sets encoder periodically inserts every IDR frame when coding；
3rd submodule, in image sets encoder, being carried out by the function for calling C++ to encapsulate in COStream
Parallel encoding.
Further, image parallel encoding module includes：
The message sub-module of CTU rows, for each two field picture in image sets to be divided into several according to HEVC standard
CTU, collects the information of CTU rows, and the information of the CTU rows includes CTU in index, the line number of CTU rows and the CTU rows of CTU rows
Number；
Image parallel encoding submodule, for the information of every row CTU rows to be inputted into a wavefront parallel encoder, is used
COStream splitjoin structures carry out parallel encoding；In addition to the first row CTU rows, remaining CTU row is intended to progress etc.
Treat that the notice until receiving a upper CTU row can just be continued executing with；Current wavefront coded device often encodes one CTU of completion just
Send and notify to next CTU rows.
Further, the parallel entropy code module of syntactic element includes：
Entropy code function submodule, several syntactic elements are included for each CTU, and syntactic element is converted into binary system ratio
Arithmetic coding is carried out after spy's string, one context probability modeling of each syntactic element correspondence, entropy code function uses arithmetic coding
And context probability modeling；
Parallel entropy code submodule, each grammer member is called for each syntactic element using COStream pipeline
The entropy code function of element, obtains entropy code streamline, and parallel entropy code is carried out to syntactic element using entropy code streamline.
In general, by the contemplated above technical scheme of the present invention compared with prior art, with following beneficial effect
Really：
(1) image sets are carried out parallel encoding by the characteristics of present invention can be handled using image sets with parallelization, accelerate coding
Speed, the characteristics of can be with parallel processing using CTU rows in image, parallel encoding be carried out to CTU rows, accelerates coding rate, is utilized
The characteristics of CTU syntactic element can carry out pipeline processes, realizes entropy code parallel processing, accelerates coding rate, with reference to each
The parallel encoding characteristic of individual level, realizes HEVC parallel encodings, accelerates coding rate.HEVC scramble time is reduced, is solved
The demand of magnanimity multimedia video data fast coding, reduces the compression time of video data under internet environment, for regarding
Frequency service provider's quick response client request is provided convenience, while reducing the time that user asks Video service, is improved
Efficiency.
(2) preferred, in open image sets, the frame in an image sets may be referred in an above image sets
, for open image sets, there is data dependence relation between image sets in frame.In such image sets, only
First frame of first image sets is every IDR frame.It is therefore, difficult during coding because Video coding is a kind of lossy coding
Exempt from that mistake can be introduced, due to inter prediction, mistake can constantly spread after introducing, in order to avoid the hair of such case
Raw, encoder can periodically insert every IDR frame when coding, and every IDR frame can make encoder empty reference frame lists, can keep away
Exempt from the lasting diffusion of code error.
Brief description of the drawings
Fig. 1 is a kind of flow chart of HEVC parallel encoding methods based on COStream provided in an embodiment of the present invention；
Fig. 2 is the master-plan block diagram for the HEVC parallel encoding methods based on COStream that the embodiment of the present invention 1 is provided；
Fig. 3 is the flow chart for the image sets parallel encoding that the embodiment of the present invention 1 is provided；
Fig. 4 is the flow chart for the single image coding that the embodiment of the present invention 1 is provided；
Fig. 5 is the wavefront parallel encoding schematic diagram that the embodiment of the present invention 1 is provided；
Fig. 6 is the flow chart for the wavefront parallel encoding framework that the embodiment of the present invention 1 is provided；
Fig. 7 is the flow chart for the wavefront parallel encoding details that the embodiment of the present invention 1 is provided；
Fig. 8 is the entropy code streamline schematic diagram that the embodiment of the present invention 1 is provided；
Fig. 9 is the entropy code parallel processing flow chart that the embodiment of the present invention 1 is provided；
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in each embodiment of invention described below
Not constituting conflict each other can just be mutually combined.
As shown in figure 1, a kind of HEVC parallel encoding methods based on COStream, including：
(1) image sets parallel encoding：Video sequence is divided into some image sets, the information of each image sets is obtained, will
The information of each image sets inputs corresponding image sets encoder, is carried out by the function for calling C++ to encapsulate in COStream
Parallel encoding；
(2) image parallel encoding：Each two field picture in image sets is divided into several CTU according to HEVC standard, collected
The information of CTU rows, a wavefront parallel encoder is inputted by the information of every row CTU rows, is tied using COStream splitjoin
Structure carries out parallel encoding；
(3) the parallel entropy code of syntactic element：Each CTU includes several syntactic elements, each syntactic element correspondence one
Entropy code function, each syntactic element calls the entropy code function of each syntactic element using COStream pipeline, obtains
To entropy code streamline, parallel entropy code is carried out to syntactic element using entropy code streamline；
(4) compression of video sequence：Image sets parallel encoding, image parallel encoding and syntactic element are carried out simultaneously to video sequence
After row entropy code, the video after being compressed.
Embodiment 1
As shown in Fig. 2 HEVC parallel encodings are by image sets parallel encoding, wavefront parallel encoding, entropy code parallel processing structure
Into original video sequence passes through after HEVC parallel encodings, generates bit stream file, completes coding.
(1) image sets are divided, image sets parallel encoding is carried out
Image sets are granularity most thick processing units in HEVC, and image sets are divided into two types, and one kind is closed type image
Group, another is open image sets.For closed type image sets, the frame in an image sets will not refer to another figure
Separate between frame in picture group, image sets, without data dependence relation, the referring-to relation between frame and frame exists only in figure
As the inside of group.Open image sets are different with closed type image sets, in open image sets, and the frame in an image sets can
With reference to the frame in an above image sets, for open image sets, there is data dependence relation between image sets.
In such image sets, first frame of only first image sets is every IDR frame.Because Video coding is that one kind damages volume
Code, therefore, can introduce mistake unavoidably during coding, due to inter prediction, and mistake can constantly expand after introducing
Dissipate, in order to avoid the generation of such case, image sets encoder can periodically insert every IDR frame, every IDR frame meeting when coding
Encoder is emptied reference frame lists, the lasting diffusion of code error can be avoided.The spacing of every IDR frame is the whole of gop size
Several times and it is necessary to more than or equal to the size of image sets.The step as shown in Figure 3 includes two sub-steps：
(1-1) pretreatment stage：Main function is the parameter of parsing input, and carrying out entirety to image sets encoder matches somebody with somebody
Put.The quantity of image in the ID of image sets, image sets is set, the size of each frame is calculated, calculates the start frame and knot of image sets
Beam frame etc.., it is necessary to which the information of image sets is assigned in each different image sets encoder after initialization is completed, image sets are compiled
Code device is obtained after the information of image sets, and it is parsed, image sets is then read, then image sets is compiled successively
Code.
(1-2) pipeline stage：After pretreatment stage is completed, into pipeline stage.Pipeline stage makes
Realization is mixed with C++ with COStream, splitjoin is a Data Stream Processing structure in COStream language, it is main to use
In the parallel task of processing data level.Whole parallel framework is built using COStream, and internal specific coding step uses C++
Realize, and be packaged into function there is provided be called to COStream.Pipeline stage includes three steps：Input processing, simultaneously
Row processing and output processing.
(1-2-1) input processing：The major function of input processing module is to read image sets according to the information of image sets, so
Image sets are put into queue afterwards, because input module and multiple images group encoder are required for grasping image sets queue
Make, it is therefore desirable to which mutual exclusion lock is protected to input rank, need to carry out locking operation to input lock before data are read, read
Take release input lock after finishing.An image sets are finished in each reading and are activated by an image sets encoder, allow image sets
Encoder is handled.
(1-2-2) parallel processing：Parallel work-flow module is the nucleus module of image sets parallelization processing, and it is by N number of image
Group encoder is constituted, and N is degree of parallelism, and depending on the setting of developer, the higher speed-up ratio of degree of parallelism is higher in theory, but specifically
The environment of consideration processing platform is needed when realizing, because CPU quantity and internal memory are conditional.Each image sets encoder
Represented with a GOPEncoder node.Before an image sets are encoded, locking operation, Waiting Graph first are carried out to input lock
It is ready as organizing, if input module has read image sets, then input lock will be discharged, represent that image sets are ready, that
Continue executing with, image sets encoder takes out an image sets from queue and handled., will not be first by figure when realizing
GOPEncoder is transmitted to again after being read out as group, because image sets data volume is very big, figure is relate to during transmission
As the duplication and movement of group, huge expense can be produced.Therefore the data in data flow are not image sets, but figure one by one
As the information of group.GOPEncoder is obtained traveling through image sets after image sets, and each image in image sets is encoded,
Mainly realized that specific encoding operation is as shown in Figure 4 by the function for calling C++ to encapsulate in COStream in this part.
(1-2-2) output is handled：After being completed to image sets coding, bit stream is produced, due in COStream,
Image sets are processed in parallel, and multiple images group encoder is run simultaneously, therefore the generation of bit stream is random, in order to keep away
Exempt from it is multiple cause dirty data while writing, write-in bit stream file before need to output lock lock, then bit stream is write
In file, release output lock after writing.
(2) CTU rows are divided, wavefront parallel encoding is realized
CTU is in order to more adapt to the change of image texture characteristic, with the one kind for obtaining more preferable encoding efficiency and proposing
New coding structure.One CTU includes a luminance coding tree block (Coding Tree Block, CTB) and two chroma coders
Tree block.Size common CTU is 64 × 64.CTU can continue to be divided into coding unit (Coding according to the form of quaternary tree
Unit, CU), each CU includes a luminance coding block (Coding Block, CB) and two chroma coder blocks, and CU can be with
It can continue to be divided into sub- CU downwards, therefore CU size can be from 8 × 8 to 64 × 64.CTU this flexible dividing mode
Add the complexity of calculating.In order to accelerate the processing speed to CTU, HEVC proposes wavefront parallel encoding, and it can be simultaneously
Multiple CTU rows are handled, the parallel encoding to CTU rows is realized.Wavefront parallel encoding is as shown in Figure 5.Except first CTU row it
Outside, remaining each CTU row will fall latter two CTU progress than a upper CTU row, because there is frame in HEVC
Prediction, the CTU that CTU needs to refer to directly over it and upper right side could carry out infra-frame prediction, therefore on each CTU rows will fall behind
One CTU row, two CTU progress.The flow of wavefront parallel encoding is as shown in fig. 6, contain two big steps.
(2-1) is initialized
The major function of initialization module is initialization encoder context, and the structure of the information of CTU rows is deposited in definition,
The quantity of CTU rows is calculated, related structure etc. is initialized, this part is realized using C++.
(2-1-1) defines CTURow data structures：In order to carry out parallelization processing, first definition structure to CTU rows
Body stores the information relevant with CTU rows, and each CTU is represented using CTURow, and it includes following information：Row_index tables
It is which row of image to show CTU rows, specifies the current abscissas of CTU to be processed in the picture, and current_col represents to work as
Preceding preparation will handle CTU row number, the current ordinates of CTU to be processed in the picture be specified, because image can be by more
Individual CTU is constituted, can be only by row_index and current_col if whole image can be regarded as CTU two-dimensional array
One determines the positions of CTU in the picture；Whether the CTU rows that is_active represents current are active, if current CTURow is being waited
Treat CTURow notice, then it is equal to false, otherwise equal to true；Semaphore is notified for event
Semaphore, needs to wait CTURow notification signal, if do not received, with regard to that before a CTU is handled
Enter blocked state until notified event is just continued executing with；Entropy is entropy code object, for entropy code operation,
In order to allow CTU row parallel processings, it is necessary to which allowing has the entropy code operation of Context-dependent separate between CTU rows,
Therefore, each CTU row is equipped with an entropy coder；The backup of entropy_buffer then entropy code objects, allows every
Although it is separate that individual CTU rows are owned by the entropy code operation that an entropy coder can be between above CTU rows, also result in
Entropy code operation can not make full use of existing context statistical information, in order to make full use of existing context to count
Information, each CTU rows need to replicate the entropy code context of a upper CTU row, entropy_buffer work before proceeding
With being that every coding completes to preserve entropy code context after a CTU, to facilitate the use of CTU rows below.
(2-1-2) calculates the quantity of CTU rows：The quantity of CTU rows is calculated according to the height of frame and CTU size,
CTURowNum=PictHeight ÷ CTUHeight.
(2-1-3) initializes the corresponding CTURow objects of each CTU row：Row_index is initialized as current CTU rows
Line number；Current_col is initialized as 0；Remaining is_active is initialized as false in addition to first CTU row；
For first CTU row, semaphore is initialized as the state of signal, and the corresponding semaphore of other CTU rows is initial
Turn to no signal condition.
(2-2) parallelization is handled：Parallel processing module realized using COStream splitjoin structures,
Processing of the splitjoin data flow architectures dedicated for data level parallel, its each node is that a wavefront is compiled parallel
Code device WPPEncoder, the number of node illustrates degree of parallelism.The unit of WPPEncoder processing is CTU rows, and it travels through CTU rows
In each CTU, complete encoding operation is carried out to each CTU, until all CTU are disposed.Due to multiple
Can be performed parallel between WPPEncoder, it is achieved that the parallelization processing of CTU rows, WPPEncoder degree of parallelism N by
It can be configured as needed.The each WPPEncoder processing to CTU rows is as shown in Figure 7.
(2-2-1) starts WPPEncoder：According to setting above to CTURow, except first WPPEncoder it
Outside, other WPPEncoder are blocked wait, because in addition to first CTURow, other CTURow semaphore
All it is no signal condition.
(2-2-2) waits notification event：Except the first row CTU, it is other per a line CTU before a CTU is handled, on
The processing progress of one CTU row must more than current CTU rows two CTU.If image is regarded as CTU two-dimensional array, frame
[x] [y] represents the CTU of xth row y row, then have to wait frame [x-1] [y+2] place before frame [x] [y] processing
Reason is completed, and otherwise can not be handled.The notice of event can be realized by semaphore, but COStream does not have the branch of semaphore
Hold, it is therefore desirable to realize that the encapsulation to semaphore operations is called there is provided interface to COStream using C++, COStream is supported
With C++ hybrid programming.If WPPEncoder have received the notice of signal, then continue to perform downwards.
(2-2-3) replicates the entropy code context of a upper CTU row：The purpose for replicating entropy code context is in order to effective
The already present context probabilistic information of utilization.Only need to replicate before CTU rows start once, behind to the institutes of the CTU rows
There is entropy code operation to be all based on this entropy code context, it is no longer necessary to replicate.
(2-2-4) CTU is encoded：The step contains the operation such as complete prediction, conversion, quantization, entropy code.This part
Directly invoke the realization of C++ functions.
(2-2-5) counts the number for the CTU that completion has been handled in current CTU rows：Count the CTU handled
The reason for number is that is, current line is than next line multiprocessing in order to calculate the processing " spacing " between next line CTU and current CTU rows
CTU number, send event for the CTU that decides whether to next line and notify.
(2-2-6) judges whether to need to send event notice：Calculate the current completed CTU of CTU rows number with it is next
The difference of the individual completed CTU of CTU rows number, if difference is more than or equal to 2, then event notice is just sent to next line CTU,
Next line is told to continue with CTU.Because the complexity of the texture of image parts is different, between CTU rows
Inconsistent situation occurs in processing progress, in order to meet the requirement of intra-prediction data dependence, it is necessary to judge whether
The condition specified is met, event is sent if meeting and is notified, is not otherwise sent.In addition, if thing can be sent by judgement
Part notifies that the next CTU rows of expression, which are in, enlivens shape, it is necessary to which next CTURow is_active fields are set into true
State.
(2-2-7) sets the state of current CTU rows：Calculate a completed CTU of CTU rows number and current CTU
The difference of the completed CTU of row number, if difference is less than 2, then representing next CTU of current CTU rows will be discontented with
The data dependence relation of sufficient infra-frame prediction is, it is necessary to which the is_active for setting current CTURow is false, and the current CTU of expression is about to
Into blocked state.
(2-2-8) judges whether current CTU rows have handled completion：If having handled completion so to exit
WPPEncoder encoders, otherwise continue cycling through execution.
(3) entropy code streamline is constructed, entropy code parallel processing is realized
Main binary arithmetic coding (the Context-based Adaptive using context-adaptive in HEVC
Binary Arithmetic Coding, CABAC) it is used as entropy code algorithm.CABAC is realized based on arithmetic coding, except ginseng
Outside manifold and slice header informations, the entropy code of remaining syntactic element and conversion coefficient is all handled using CABAC.It
It is combined using arithmetic coding and context probability modeling, gives an efficient entropy coding method.Context probability mould
Type is that by well-designed, the probability of symbol is obtained by tabling look-up, and needs to enter probabilistic model after completing symbolic coding
Row updates, therefore, and the probability of current sign depends on the result of above entropy code.The statistics that CABAC takes full advantage of symbol is special
Property and context result.
CABAC is Context-dependent, and when carrying out arithmetic coding, the probability of symbol comes from context model, is compiled
The complete symbol of code needs to update context model, and therefore, CABAC has data dependence relation.CABAC context model with
Syntactic element is related, syntactic element one context probability modeling of correspondence, just right after context model initialization is completed
All syntactic elements are handled in CTU, until CTU all in slice is disposed, then in next slice
Original position entropy coder is reset, the main function of replacement is to rebuild and initialize context probability modeling.
, it is necessary to encode CTU all in slice in sequence when carrying out entropy code to a slice, for single CTU, its inside
Syntactic element is also required to be encoded in sequence, therefore, and the parallel characteristics of data level are not present for slice inside.
The characteristics of according to context probability modeling, is analyzed, and finds in same slice, the renewal of context probability modeling with
Based on syntactic element, identical syntactic element uses same context mould, and different syntactic elements are used above and below different
Literary probabilistic model, that is to say, that for different syntactic elements, their context probability modeling is independent, before CTU is different
Put, different syntactic elements can be encoded simultaneously.Therefore, although parallel characteristics of the CTU entropy code without data level,
But possessing the parallel characteristics of task level, the parallel characteristics of task level are also streamline.CTU entropy code flow according to grammer
The coded sequence of element is divided into multiple steps, forms a streamline, and each step corresponds to a stage of streamline, leads to
The mode for crossing streamline is realized and the parallelization of entropy code can be handled.Slice entropy code streamline is as shown in figure 8, CTU is pressed
Performed all coding flows according to the coded sequence of syntactic element, different CTU can be handled simultaneously, with CTU1 and
Exemplified by CTU2, CTU1 skip_flag and CTU2 split_flag can be encoded simultaneously, because the two syntactic elements
Context probability modeling it is separate, without data dependence, and not in same CTU, it is possible to simultaneously encode.Entropy
The flow for encoding parallel processing is as shown in Figure 9.
(3-1) pretreatment module：TU is handled by defining a recursive function, current TU is recorded using TUInfo objects
Information, judge whether TU may proceed to downwards according to the model split of quaternary tree for sub- TU, if current TU continues division downwards,
So represent that current TU does not need encoding transform coefficients and other syntactic elements related to conversion coefficient, TUInfo is_
Split is set to true, if TU does not continue to be divided downwards, then need to encode all including conversion coefficient
Syntactic element, TUInfo is_split fields are set to false, then record relative indexings of the current TU in CTU and
Depth in CTU, is then added it in queue, and queue directly uses the dique in C++ standard template libraries.Due to
Orders of the TUInfo in queue is exactly the order that TU is encoded, therefore entropy code module is to TUInfo sequential processes and recurrence
The order that TU is handled under mode is identical, does not result in confusion.
(3-2) entropy code parallelization module：COStream pipeline data flow architectures dedicated for processing streamline,
Support the parallel of task level., each language in the stage of the coding of each syntactic element in TU as streamline
The calculate node actor, TU that the encoding operation of method is packaged into pipeline often just encode a language by a calculate node
Method element, after TU is by pipeline, is just completed to the entropy code of the TU.One TUInfo queue is comprising multiple
TUInfo object, each object represents a TU, for any one TUInfo object, and the calculate node in pipeline is all
Judge the value of is_split attributes, if true, represent that the TU that the TUInfo is represented may proceed to be divided into sub- TU downwards,
So only need to encode split_flag syntactic elements, other syntactic elements, which include conversion coefficient, to be encoded, such as
Fruit is_split is false, represents that the TU that TUInfo is represented does not continue to downward division, then needing to encode includes split_
All syntactic elements including flag.The entropy code of various syntactic elements mainly uses CABAC as realization in HEVC, main
To include three steps：Binarization, context modeling, binary arithmetic coding.
(3-2-1) binarization：The object of CABAC arithmetic coding processing is binary Bit String, therefore, in arithmetic
The syntactic element in HEVC is needed to be converted into binary bits string before coding.Common binarizing method has：Unitary code,
Block unitary code, block Lay this code, k ranks Exp- Golomb, fixed length code etc..The binarizing method of each syntactic element is not to the utmost
Identical, the corresponding binarizing method of specific syntactic element can check HEVC standard document.
(3-2-2) context modeling：In HEVC, there is correlation between syntactic element, the probability of current sign according to
The result of entropy code before Lai Yu, current sign can be more accurately estimated using the information of above encoded symbol
Probability, these encoded information are thus referred to as context.In general, the probability of symbol becomes with the content of video
Change.A suitable contextual information is selected to be exactly context modeling for syntactic element.HEVC standard document gives all
Syntactic element initial context, the context of different syntactic elements is typically different.
(3-2-3) binary arithmetic coding：After the corresponding context model initialization of syntactic element is completed, proceed by
Binary arithmetic coding.Binary arithmetic coding is to carry out arithmetic coding to the corresponding binary bits string of syntactic element.Two enter
Arithmetic coding processed includes two ways：Bypass mode and usual manner.Bypass coded system in, the symbol of binary string it is general
Rate is identical, it is not required that update context probability modeling；In conventional manner, the probability of symbol can be by binary string
Obtained in context model, be required for being updated context model after completing each symbolic coding.Use routine
Mode or bypass mode determine by syntactic element, HEVC documents specify which syntactic element using bypass mode which
Syntactic element uses usual manner.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, it is not used to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the invention etc., it all should include
Within protection scope of the present invention.
Claims (8)
1. a kind of HEVC parallel encoding methods based on COStream, it is characterised in that including：
(1) image sets parallel encoding：Video sequence is divided into some image sets, the information of each image sets is obtained, will be each
The information of image sets inputs corresponding image sets encoder, parallel by calling the function that C++ is encapsulated to carry out in COStream
Coding；
(2) image parallel encoding：Each two field picture in image sets is divided into several CTU according to HEVC standard, CTU rows are collected
Information, the information of every row CTU rows is inputted into a wavefront parallel encoder, entered using COStream splitjoin structures
Row parallel encoding；
(3) the parallel entropy code of syntactic element：Each CTU includes several syntactic elements, and one entropy of each syntactic element correspondence is compiled
Code function, each syntactic element is called the entropy code function of each syntactic element using COStream pipeline, obtains entropy
Encoding pipeline, parallel entropy code is carried out using entropy code streamline to syntactic element；
(4) compression of video sequence：Image sets parallel encoding, image parallel encoding and the parallel entropy of syntactic element are carried out to video sequence
After coding, the video after being compressed.
2. a kind of HEVC parallel encoding methods based on COStream according to claim 1, it is characterised in that the step
Suddenly (1) includes：
Video sequence is divided into some image sets by (1-1), obtains the information of each image sets, and the information of described image group includes
The size of each two field picture, the start frame of image sets and end in the quantity of image, image sets in the ID of image sets, image sets
Frame；
The information of each image sets is inputted corresponding image sets encoder by (1-2), right when image sets are open image sets
The image sets encoder answered periodically inserts every IDR frame when coding；
(1-3) carries out parallel encoding in image sets encoder by the function for calling C++ to encapsulate in COStream.
3. a kind of HEVC parallel encoding methods based on COStream according to claim 1, it is characterised in that the step
Suddenly (2) include：
Each two field picture in image sets is divided into several CTU by (2-1) according to HEVC standard, collects the information of CTU rows, described
CTU number in index of the information of CTU rows including CTU rows, the line number of CTU rows and CTU rows；
The information of every row CTU rows is inputted a wavefront parallel encoder by (2-2), uses COStream splitjoin structures
Carry out parallel encoding；In addition to the first row CTU rows, remaining CTU row is intended to progress and waited until receive a CTU row
Notice, can just continue executing with；Current wavefront coded device often encodes one CTU of completion and just sends notice to next CTU rows.
4. the HEVC parallel encoding methods according to claim 1 based on COStream, it is characterised in that the step
(3) include：
(3-1) each CTU includes several syntactic elements, and syntactic element is converted into after binary bits string carrying out arithmetic coding,
One context probability modeling of each syntactic element correspondence, entropy code function uses arithmetic coding and context probability modeling；
(3-2) each syntactic element calls the entropy code function of each syntactic element using COStream pipeline, obtains
Entropy code streamline, parallel entropy code is carried out using entropy code streamline to syntactic element.
5. a kind of HEVC parallel encoder systems based on COStream, it is characterised in that including：
Image sets parallel encoding module, for video sequence to be divided into some image sets, obtains the information of each image sets, will
The information of each image sets inputs corresponding image sets encoder, is carried out by the function for calling C++ to encapsulate in COStream
Parallel encoding；
Image parallel encoding module, for each two field picture in image sets to be divided into several CTU according to HEVC standard, is collected
The information of CTU rows, a wavefront parallel encoder is inputted by the information of every row CTU rows, is tied using COStream splitjoin
Structure carries out parallel encoding；
The parallel entropy code module of syntactic element, several syntactic elements, each syntactic element correspondence one are included for each CTU
Entropy code function, each syntactic element calls the entropy code function of each syntactic element using COStream pipeline, obtains
To entropy code streamline, parallel entropy code is carried out to syntactic element using entropy code streamline；
Compression of video sequence module, for carrying out image sets parallel encoding, image parallel encoding and syntactic element to video sequence
After parallel entropy code, the video after being compressed.
6. a kind of HEVC parallel encoder systems based on COStream according to claim 5, it is characterised in that the figure
As group parallel encoding module includes：
First submodule, for video sequence to be divided into some image sets, obtains the information of each image sets, described image group
The ID of information including image sets, in image sets in the quantity of image, image sets the size, image sets of each two field picture starting
Frame and end frame；
Second submodule, for the information of each image sets to be inputted into corresponding image sets encoder, when image sets are open
During image sets, corresponding image sets encoder periodically inserts every IDR frame when coding；
3rd submodule, it is parallel by calling the function that C++ is encapsulated to carry out in COStream in image sets encoder
Coding.
7. a kind of HEVC parallel encoder systems based on COStream according to claim 5, it is characterised in that the figure
As parallel encoding module includes：
The message sub-module of CTU rows, for each two field picture in image sets to be divided into several CTU according to HEVC standard, is received
Collect the information of CTU rows, the information of the CTU rows includes the number of CTU in index, the line number of CTU rows and the CTU rows of CTU rows；
Image parallel encoding submodule, for the information of every row CTU rows to be inputted into a wavefront parallel encoder, is used
COStream splitjoin structures carry out parallel encoding；In addition to the first row CTU rows, remaining CTU row is intended to progress etc.
Treat that the notice until receiving a upper CTU row can just be continued executing with；Current wavefront coded device often encodes one CTU of completion just
Send and notify to next CTU rows.
8. the HEVC parallel encoder systems according to claim 5 based on COStream, it is characterised in that the grammer member
Plain parallel entropy code module includes：
Entropy code function submodule, several syntactic elements are included for each CTU, and syntactic element is converted into binary bits string
After carry out arithmetic coding, one context probability modeling of each syntactic element correspondence, entropy code function uses arithmetic coding and upper
Hereafter probabilistic model；
Parallel entropy code submodule, each syntactic element is called for each syntactic element using COStream pipeline
Entropy code function, obtains entropy code streamline, and parallel entropy code is carried out to syntactic element using entropy code streamline.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710477885.8A CN107197296B (en) | 2017-06-22 | 2017-06-22 | A kind of HEVC parallel encoding method and system based on COStream |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710477885.8A CN107197296B (en) | 2017-06-22 | 2017-06-22 | A kind of HEVC parallel encoding method and system based on COStream |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107197296A true CN107197296A (en) | 2017-09-22 |
| CN107197296B CN107197296B (en) | 2019-08-13 |
Family
ID=59879428
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710477885.8A Active CN107197296B (en) | 2017-06-22 | 2017-06-22 | A kind of HEVC parallel encoding method and system based on COStream |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107197296B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113196768A (en) * | 2018-12-21 | 2021-07-30 | 索尼集团公司 | Image processing apparatus and method |
| CN113557731A (en) * | 2019-03-11 | 2021-10-26 | 佳能株式会社 | Method, apparatus and system for encoding and decoding a block tree of video samples |
| CN114257839A (en) * | 2020-09-23 | 2022-03-29 | 京东方科技集团股份有限公司 | Video encoding device, video decoding device, video playing system and video playing method |
| WO2024021772A1 (en) * | 2022-07-29 | 2024-02-01 | 腾讯科技（深圳）有限公司 | Live streaming media data processing method, system and apparatus, and computer device |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103974081A (en) * | 2014-05-08 | 2014-08-06 | 杭州同尊信息技术有限公司 | HEVC coding method based on multi-core processor Tilera |
| CN104221381A (en) * | 2012-04-11 | 2014-12-17 | 高通股份有限公司 | Wavefront parallel processing for video coding |
| CN104488267A (en) * | 2012-06-29 | 2015-04-01 | 高通股份有限公司 | Tiles and wavefront parallel processing |
| CN104811696A (en) * | 2015-04-17 | 2015-07-29 | 北京奇艺世纪科技有限公司 | Video data coding method and device |
| CN104869398A (en) * | 2015-05-21 | 2015-08-26 | 大连理工大学 | Parallel method of realizing CABAC in HEVC based on CPU+GPU heterogeneous platform |
| CN104980764A (en) * | 2014-04-14 | 2015-10-14 | 深圳中兴力维技术有限公司 | Parallel coding/decoding method, device and system based on complexity balance |
| CN105516728A (en) * | 2015-12-15 | 2016-04-20 | 华中科技大学 | Parallel intra-frame prediction method of 8*8 sub-macroblocks in H.265/HEVC |
| US20160323586A1 (en) * | 2012-09-26 | 2016-11-03 | Sun Patent Trust | Image decoding method, image coding method, image decoding apparatus, image coding apparatus, and image coding and decoding apparatus |
| US20160381365A1 (en) * | 2015-06-23 | 2016-12-29 | Magnum Semiconductor, Inc. | Apparatuses and methods for performing information extraction and insertion on bitstreams |
| CN106788467A (en) * | 2016-11-28 | 2017-05-31 | 华中科技大学 | A kind of Raptor Code coding methods based on CUDA, interpretation method and system |
-
2017
- 2017-06-22 CN CN201710477885.8A patent/CN107197296B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104221381A (en) * | 2012-04-11 | 2014-12-17 | 高通股份有限公司 | Wavefront parallel processing for video coding |
| CN104488267A (en) * | 2012-06-29 | 2015-04-01 | 高通股份有限公司 | Tiles and wavefront parallel processing |
| US20160323586A1 (en) * | 2012-09-26 | 2016-11-03 | Sun Patent Trust | Image decoding method, image coding method, image decoding apparatus, image coding apparatus, and image coding and decoding apparatus |
| CN104980764A (en) * | 2014-04-14 | 2015-10-14 | 深圳中兴力维技术有限公司 | Parallel coding/decoding method, device and system based on complexity balance |
| CN103974081A (en) * | 2014-05-08 | 2014-08-06 | 杭州同尊信息技术有限公司 | HEVC coding method based on multi-core processor Tilera |
| CN104811696A (en) * | 2015-04-17 | 2015-07-29 | 北京奇艺世纪科技有限公司 | Video data coding method and device |
| CN104869398A (en) * | 2015-05-21 | 2015-08-26 | 大连理工大学 | Parallel method of realizing CABAC in HEVC based on CPU+GPU heterogeneous platform |
| US20160381365A1 (en) * | 2015-06-23 | 2016-12-29 | Magnum Semiconductor, Inc. | Apparatuses and methods for performing information extraction and insertion on bitstreams |
| CN105516728A (en) * | 2015-12-15 | 2016-04-20 | 华中科技大学 | Parallel intra-frame prediction method of 8*8 sub-macroblocks in H.265/HEVC |
| CN106788467A (en) * | 2016-11-28 | 2017-05-31 | 华中科技大学 | A kind of Raptor Code coding methods based on CUDA, interpretation method and system |
Non-Patent Citations (3)
| Title |
|---|
| YU-HSIN CHEN，ET AL.: "A Deeply Pipelined CABAC Decoder for HEVC Supporting Level 6.2 High-Tier Applications", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
| 张维维，等.: "COStream:一种面向数据流的编程语言和编译器实现", 《计算机学报》 * |
| 杨秋吉,等.: "面向Storm的数据流编程模型与编译优化方法研究", 《计算机工程与科学》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113196768A (en) * | 2018-12-21 | 2021-07-30 | 索尼集团公司 | Image processing apparatus and method |
| CN113557731A (en) * | 2019-03-11 | 2021-10-26 | 佳能株式会社 | Method, apparatus and system for encoding and decoding a block tree of video samples |
| CN113557731B (en) * | 2019-03-11 | 2024-04-02 | 佳能株式会社 | Method, apparatus and system for encoding and decoding a block tree of video samples |
| US11949857B2 (en) | 2019-03-11 | 2024-04-02 | Canon Kabushiki Kaisha | Method, apparatus and system for encoding and decoding a tree of blocks of video samples |
| CN114257839A (en) * | 2020-09-23 | 2022-03-29 | 京东方科技集团股份有限公司 | Video encoding device, video decoding device, video playing system and video playing method |
| WO2024021772A1 (en) * | 2022-07-29 | 2024-02-01 | 腾讯科技（深圳）有限公司 | Live streaming media data processing method, system and apparatus, and computer device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107197296B (en) | 2019-08-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| USRE49727E1 (en) | System and method for decoding using parallel processing | |
| JP7098805B2 (en) | Codec using neural network | |
| CN107197296B (en) | A kind of HEVC parallel encoding method and system based on COStream | |
| CN103918273B (en) | It is determined that the method for the binary code word for conversion coefficient | |
| TWI239207B (en) | Method and device for coding digital images using intra-mode block prediction | |
| CN102150428B (en) | System and method for video encoding using adaptive segmentation | |
| CN101616323B (en) | System and method for decoding video coding data stream | |
| CN105453570A (en) | Content adaptive entropy coding of partitions data for next generation video | |
| CN103931197A (en) | Method of determining binary codewords for transform coefficients | |
| CN105120277A (en) | Method for encoding and decoding images, encoding and decoding device, and corresponding computer programs | |
| US10455244B2 (en) | Method and device for entropy encoding or entropy decoding video signal for high-capacity parallel processing | |
| CN110337002B (en) | HEVC (high efficiency video coding) multi-level parallel decoding method on multi-core processor platform | |
| US20120155540A1 (en) | Pixel retrieval for frame reconstruction | |
| CN106358043A (en) | Hybrid video decoding apparatus and associated hybrid video decoding method | |
| CN103947211A (en) | Method and apparatus of video encoding with partitioned bitstream | |
| CN106412597A (en) | Work load balancing in multicore video decoder | |
| US9706201B2 (en) | Region-based processing of predicted pixels | |
| Jilani et al. | JPEG image compression using FPGA with Artificial Neural Networks | |
| CN115866245A (en) | Video encoding method, video encoding device, computer equipment and storage medium | |
| CN116016951A (en) | Point cloud processing method, device, equipment and storage medium | |
| US20180316920A1 (en) | Video image encoding device, video image encoding method and program recording medium | |
| CN112437303B (en) | JPEG decoding method and device | |
| CN109672889A (en) | The method and device of the sequence data head of constraint | |
| EP4391538A1 (en) | Point cloud coding method and apparatus, point cloud decoding method and apparatus, and computer and storage medium | |
| CN113259675B (en) | Ultrahigh-definition video image parallel processing 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |