CN103414895A - Encoder intra-frame prediction device and method applicable to HEVC standards - Google Patents
Encoder intra-frame prediction device and method applicable to HEVC standards Download PDFInfo
- Publication number
- CN103414895A CN103414895A CN2013103216770A CN201310321677A CN103414895A CN 103414895 A CN103414895 A CN 103414895A CN 2013103216770 A CN2013103216770 A CN 2013103216770A CN 201310321677 A CN201310321677 A CN 201310321677A CN 103414895 A CN103414895 A CN 103414895A
- Authority
- CN
- China
- Prior art keywords
- prediction
- reference pixel
- order
- pixel
- scanning
- 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.)
- Pending
Links
Images
Landscapes
- Compression, Expansion, Code Conversion, And Decoders (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention belongs to the technical field of high-definition digital video compression coding and decoding, and particularly relates to an encoder intra-frame prediction device and method applicable to HEVC standards. The encoder intra-frame prediction device comprises a control unit, two prediction engines, a rebuilt loop and a series of buffers. One prediction engine (A) is arranged at the first stage of a system stage assembly line and is responsible for selecting the optimal prediction mode and the best block dividing situation, the other prediction engine (B) and the rebuilt loop are combined to be arranged at the second stage of the system stage assembly line, results of the prediction engine (A) are used for conducting prediction and calculating residual errors, and then conversion processing, quantization processing, inverse quantization processing and inverse conversion processing are carried out on the residual errors. According to the encoder intra-frame prediction device and method, the data dependency in the prediction process can be effectively reduced, and the throughput rate can be improved. In addition, a common prediction unit and an appropriate scanning sequence reduce the cost of hardware, an internal high-streamlined structure enables the working frequency to be remarkably improved, and accordingly real-time coding on the high-definition video is achieved.
Description
Technical field
The invention belongs to high-definition digital video compression coding and decoding technical field, for the HEVC video encoding and decoding standard, be specially a kind of infra-frame prediction device and method of the HEVC of meeting standard.
Background technology
HEVC (High Efficiency Video Coding) is combined the video encoding and decoding standard of future generation of organizing JCTVC to propose of establishment by international telecommunication tissue (ITU) and Motion Picture Experts Group (MPEG).Target is under the prerequisite of identical visual effect, and than the previous generation standard H.264/AVC, compression ratio doubles.
The basic process of video compression coding can be summarized as follows: 1. utilize certain mode to when pre-treatment image block pixel, predicting; 2. original pixel value is subtracted each other and obtains residual values with prediction pixel value out; 3. residual error is carried out to conversion and quantification treatment, the residual error coefficient that obtains exporting advanced the entropy coding again and formed last compressed bit stream; 4. residual error coefficient is processed through inverse quantization and inverse transformation, then obtains rebuilding pixel with the predict pixel addition obtained before, and storage is as the reference pixel of prediction.
Be similar to H.264/AVC, HEVC has still adopted the method for block-based coded system and predictive coding:
Processing unit piece in HEVC has been introduced the structure of quaternary tree, and the size of image processing block is 64 * 64 to the maximum, and it can also continue recursively to be divided into 32 * 32,16 * 16,8 * 8,4 * 4 fritter combination, and processes respectively.Coding side will once travel through all dividing condition of piece, to determine the best that is treated to of which kind of dividing condition.
Infra-frame prediction is a kind of predictive coding mode that effectively reduces spatial redundancies, it utilizes the similitude between the neighborhood pixels of same width image, with neighbor, the current block pixel is predicted, can effectively improve compression ratio like this, predict more accurately, compression ratio is higher.So at coding side, same processing unit is often needed first to carry out to the prediction of multiple predictive mode, the calculation rate distortion value, and therefrom select accordingly optimum pattern, namely predict the most accurate a kind of pattern.
Fig. 1 has set forth infra-frame prediction in HEVC and the detailed process of Angular prediction.1 predictive mode increases to 35 kinds (Planar, DC and 33 kinds of Angular patterns), wherein Planar and DC pattern are mainly used in the prediction of smooth homogeneous area in image, the Angular pattern is mainly to exist the zone of obvious directivity to predict to texture in those images, its forecasting process is mainly the direction according to modal representation, current pixel is projected in the row or row of reference pixel, then according to formula (1), carry out the interpolation of 1/32 pixel precision, in calculating, use design parameter and calculated by formula (2) (3).
Wherein A, B mean reference pixel, and P means predict pixel, and A, B select according to iIdx, and x means the abscissa (calculate with x in pattern 2 to 17 o'clock formula (2) (3), 18 to 34 o'clock x of pattern will substitute with the ordinate y of current pixel) of current pixel.
The reference pixel of using is the neighbor that is positioned at current prediction piece top, left, upper left side, upper right side and lower left, in HEVC, reference pixel has also been carried out to some special processing, mainly contain to the supposition of non-existent pixel fill, the projection of reference pixel between the filtering processing relevant to pattern and block size and ranks.
In intra-framed prediction module, the pixel that just will use peripheral piece due to the prediction of current block as a reference, is difficult to exist serious data dependency with avoiding in forecasting process, and this also becomes and hinders the bottleneck that improves throughput.In addition, increasing with the variation of block size of schema category all improved the complexity that hardware calculates, and too throughput caused to no small impact.
Summary of the invention
In order to overcome the deficiencies in the prior art, the present invention proposes a kind of interior prediction unit of the encoder frames that is applicable to the HEVC standard and method of high-throughput.
The present invention proposes a kind of interior prediction unit of encoder frames of the HEVC of being applicable to standard, it comprises a control unit, two prediction engine (A and B), a reconstruction loop and the buffer that some are essential.
Control unit, while in order to be controlled at, encoding, 35 kinds of traversal order that predictive mode is predicted respectively that traversal order and the employing standard of the quad-tree partition situation of basic processing unit piece proposed, in addition also in order to control in some forecasting processes the stand-by period to process of reconstruction, to meet the sequential requirement.
Above-mentioned control unit comprises a simple state machine and a plurality of counter; Described simple state machine, be used to controlling 4 * 4,8 * 8,16 * 16,32 * 32 forecasting process, the method scanning quad-tree structure of employing postorder traversal; Described counter, for controlling the forecasting process of different masses size, the order of mode scanning and the order of processing fritter in bulk.
Prediction engine A comprises a reference pixel processing unit, a general predicting unit and a judgement selected cell; Prediction engine B comprises a reference pixel processing unit, a general predicting unit and a residual computations unit,
Described reference pixel processing unit, process in order to filling that reference pixel is derived, filtering; It comprises 129 registers, in order to the stored reference pixel, uses 17 when carrying out 4 * 4 predictions, uses 33 while carrying out 8 * 8 predictions, uses 65 while carrying out 16 * 16 predictions, uses 129 while carrying out 32 * 32 predictions;
Described general predicting unit, be three grades of fully pipelined architectures of 16 pixel parallel processings, can support the forecasting process of all 35 kinds of patterns of 4 * 4,8 * 8,16 * 16,32 * 32 block sizes;
Described judgement selected cell, the SATD value of working as 4 * 4 of pre-treatments by calculating, namely by predicted value and original value, subtract each other, difference is done to the Hadamard conversion, relatively choose the predictive mode of the conduct optimum that the SATD value is less and best piece division by corresponding cumulative sum;
Described residual computations unit, subtract each other with the pixel that prediction obtains the residual values obtained be used to calculating original pixels.
Prediction engine A is first usingd original pixels as the reference pixel, the processes such as filling, filtering of deriving by reference to pixel processing unit, the reference pixel of really using is provided for predicting unit, thereby predict, judge that the SATD value that selected cell is predicted by calculating judges the quality of predictive mode and piece dividing condition, thereby select optimum predictive mode and best piece dividing condition.
Similar in the work of the reference pixel processing unit that prediction engine B comprises and general predicting unit and prediction engine A, but that to rebuild pixel be reference pixel, the scanning process that prediction engine B does not predict, quality without judgment model and piece division, directly utilize optimal prediction modes and the optimical block dividing condition that prediction engine A obtains to predict, finally original pixels and the pixel that obtains of prediction are subtracted each other and obtain residual values and pass to and rebuild loop.Rebuild loop residual values is carried out conversion (DCT/DST), quantized, the residual error coefficient that obtains exporting, then pass through inverse quantization and inverse transformation (IDCT/IDST) obtains rebuilding pixel, B provides reference pixel for prediction engine.
The general predicting unit comprised in prediction engine A and B is three grades of fully pipelined architectures, as shown in Figure 4, can support the prediction of 35 kinds of patterns of 4 * 4,8 * 8,16 * 16,32 * 32 size block, each 16 pixels of parallel processing (i.e. piece of 4 * 4 sizes, mean that 8 * 8 split into 44 * 4 and predict, 16 * 16 split into 16 4 * 4 and predict, 32 * 32 split into 64 4 * 4 and predict).Predicting unit only has three simple control signals: characterize " Mode " signal of the predictive mode of current employing, characterize " CU_size " signal of current prediction block sizes and characterize 4 * 4 " Blk4 * 4_num " signals that are arranged in the position of basic processing unit piece when pre-treatment.
The streamline first order comprises a reference pixel projection and initial option unit, and reference pixel is carried out to preliminary processing and selection; Coordinate computing unit and the Prediction Parameters computing unit of one current 4 * 4, in order to calculate according to control signal the parameter that forecasting process need to be used; A pre-measuring angle look-up table, in order to therefrom to search angle value to calculate relevant parameter, i.e. iIdx and iFact in formula (2) (3) according to current predictive mode.
The second level of streamline comprises four reference pixel selectors, and the parameter of calculating according to previous stage, respectively to current place
Manage each row or each row of 4 * 4, select corresponding reference pixel;
The third level of streamline comprises four interpolation devices, in order to the reference pixel that utilizes previous stage to obtain, respectively to working as pre-treatment 4 * 4
Each row of piece or each row carry out the process of interpolative prediction.
The present invention also provides a kind of intra-frame prediction method of the HEVC of being applicable to standard, mainly comprises the following steps:
(1) at first carry out the scanning traversal prediction (original pixels of take is reference pixel) of basic processing unit predictive mode and piece dividing condition, and therefrom select optimum predictive mode and best piece dividing condition;
(2) carry out accordingly again the process of second prediction (the reconstruction pixel of take is reference pixel), obtain final predicted value, calculate afterwards residual error, and it is carried out to the DST(prediction block sizes is 4 * 4) or the DCT(prediction block sizes be 8 * 8,16 * 16,32 * 32) change quantization, the residual error coefficient that obtains exporting, carry out inverse quantization, inverse transformation (IDCT/IDST), obtain rebuilding pixel value again.
In above-mentioned steps (1), the SATD value that the Hadamard conversion of calculating 4 * 4 of take while selecting optimum predictive mode and best piece dividing condition obtains is benchmark, the SATD value sum of four 4 * 4 is the SATD value of 8 * 8, in like manner obtain 16 * 16 and 32 * 32 SATD value, SATD value smaller is optimal situation.
Above-mentioned scanning in the traversal forecasting process, the scanning of dividing for the block structure of quaternary tree, what adopt is the method for postorder traversal, namely from " leaf " of lowermost layer, start to operate, handle four " leaves " operates their " root " at every turn, continue by that analogy, each like this handle four " leaves " and one " root ", just can from two kinds, select one as best dividing condition, another thoroughly to be abandoned, need not store, thereby save the area of intermediate variable memory; Concrete step is:
(1) from the upper left side of basic processing unit piece, first to first 4 * 4, adopt 35 kinds of predictive modes to predict,
According to the order of Z scanning, successively its right, 35 kinds of predictive modes of three 4 * 4 the employings following and limit, bottom right are predicted again;
(2) after finishing the prediction of four 4 * 4,8 * 8 of this four 4 * 4 compositions are carried out to the prediction that prediction block sizes is 35 kinds of predictive modes of 8 * 8;
(3 finish this prediction of 8 * 8 after, according to the order of Z scanning, successively to its right, below and the limit, bottom right
Three 8 * 8 processing procedures of carrying out step (1), (2);
(4 finish the prediction of four 8 * 8 after, 16 * 16 of this four 8 * 8 compositions carried out to prediction block sizes be
The prediction of 35 kinds of predictive modes of 16 * 16;
(5 finish this prediction of 16 * 16 after, according to the order of Z scanning, successively to its right, below and the limit, bottom right
Three 16 * 16 processing procedures of carrying out step (1) (2) (3) (4);
(6) after finishing the prediction of four 16 * 16,32 * 32 of this four 16 * 16 compositions are predicted to piece is large
Little is the prediction of 35 kinds of predictive modes of 32 * 32.
Above-mentioned scanning in the traversal forecasting process, for the scanning of 35 kinds of predictive modes, concrete step is:
(1) for the scanning of 4 * 4, because the parallel processing element of forecasting process is 16 pixels, namely one 4 * 4,
While like this pattern being scanned, the prediction that only need to carry out successively 35 kinds of patterns gets final product;
(2) for the scanning of bulk, take and 8 * 8 be example, first with a kind of predictive mode, its four 4 * 4 of comprising are advanced respectively
The row prediction, and then change another kind of pattern and equally four 4 * 4 are predicted, until traveled through all patterns; 16 * 16 and 32 * 32 adopt similar method.
Before above-mentioned each forecasting process, need the processing procedure to derive filling, the filtering of reference pixel, so that the finally all reference pixel values for predicting to be provided; Detailed process comprises: which reference pixel (1) judgement periphery exists, and which reference pixel does not exist; (2) utilize the reference pixel existed to derive to non-existent reference pixel, be filled into the position of non-existent reference pixel; (3) reference pixel is carried out to the filtering processing.
Intra-frame prediction method of the present invention, be responsible for all situations is scanned the prediction of traversal, and the original pixels of take is reference pixel, to reduce data dependence, need not wait for that the process of reconstruction of current block is finished the prediction that just can do next piece; Be responsible for the prediction of adopting optimal prediction modes and optimical block dividing condition to carry out, the reconstruction pixel of take is reference pixel, the process of reconstruction that needs to wait for current processing block before next piece is predicted is finished, and to guarantee the unification of coding side and decoding end, guarantees the quality of coding.
Intra-frame prediction method of the present invention, two forecasting processes are placed on respectively to the two-stage of system flow waterline, the scanning ergodic process of prediction is the first order, the prediction that to carry out optimization model and optimical block dividing condition with rebuild loop and be placed on the second level, suppose that basic processing unit is 32 * 32: the first order scans the process of traversal, according to 16 pixels of a parallel processing, need to be to prediction block sizes 4 * 4, 8 * 8, 16 * 16 and 32 * 32 4 kinds of dividing condition are predicted, every kind all needs to do 4 * 4, (32 * 32)/(4 * 4)=64, each 4 * 4 will be with 35 kinds of predictive modes, so need to do altogether 4 * 64 * 35=8960 time prediction, only predict for optimization model and optimum division situation the second level, namely only need to adopt a kind of pattern to predict to 64 4 * 4, namely only need to do 64 predictions altogether, although so the time that the second level need to be grown wait for completing of process of reconstruction, but still the time match of energy and the first order, form better pipeline organization, thereby improve whole throughput.
In the present invention, prediction engine A adopts original pixels as the reference pixel, although original pixels exists all the time, need not wait for that process of reconstruction completes, but still to suppose that it need to wait for that process of reconstruction judges that accordingly which reference pixel of periphery exists, which reference pixel does not exist, and utilizes the reference pixel existed to derive to non-existent reference pixel, be filled into the position of non-existent reference pixel, improved coding quality.
The present invention supports the prediction of all predictive mode and all block sizes, is easy to control, to facilitate the scanning process of coding side; The present invention adopts suitable scanning sequency, has effectively saved size and the hardware spending of memory during hardware is realized.
The accompanying drawing explanation
Fig. 1: HEVC infra-frame prediction and Angular prediction detailed process.
Fig. 2: prediction unit in the HEVC encoder frames.
Fig. 3: prediction scheme system works sequential chart in the HEVC encoder frames.
Fig. 4: the general predicting unit framework of full flowing water.
Fig. 5: the postorder scanning of quad-tree structure.
Fig. 6: the sequential chart of mode scanning and quaternary tree scanning.
Fig. 7: size is 32 * 32 basic processing unit piece.
Embodiment
The present invention will be further described below in conjunction with the drawings and specific embodiments.
Be illustrated in figure 2 a kind of interior prediction unit of encoder frames of the HEVC of being applicable to standard, comprise a control unit, two prediction engine (A and B), a reconstruction loop and the buffer that some are essential.
Use prediction unit in encoder frames provided by the invention, the detailed process of predicting is as follows:
1. the workflow of prediction engine A
First according to control unit, pass the pattern of coming, prediction block sizes and positional information, read the reference pixel that corresponding original pixels is used as prediction, and to its filling of deriving, filtering is processed, the reference pixel that obtains final output is delivered in the general predicting unit of three grades of flowing water, predicting unit is predicted accordingly and is obtained predict pixel, finally according to original pixels and predict pixel, calculate SATD, namely first subtract each other and obtain difference, again difference is carried out to the Hadamard conversion, carry out model selection and block-size selection: for model selection, get predictive mode that the SATD value is little as more excellent pattern, for block-size selection, same, the SATD sum when several fritters that the SATD while relatively carrying out the optimization model prediction as bulk and it comprise utilize optimization model to predict, the former little bulk of selecting, the little division of selecting it is carried out fritter of the latter.
2. the workflow of prediction engine B
According to the pattern from reading middle buffer memory, prediction block sizes and positional information, read the reference pixel that corresponding reconstruction pixel is used as prediction, and its filling of deriving, filtering are processed, the reference pixel that obtains final output is delivered in the general predicting unit of three grades of flowing water, predicting unit is predicted accordingly and is obtained predict pixel, and deposit buffer memory in, then original pixels and predict pixel are subtracted each other to the residual error obtained be delivered to and rebuild loop.
3. rebuild the workflow in loop
It is 4x4 that the residual error data obtained is carried out to the DST(prediction block sizes) or the DCT(prediction block sizes be 8x8,16x16,32x32) conversion, then quantize, obtain the coefficient matrix of final output, then carry out inverse quantization, inverse transformation, again with buffer memory in the predict pixel addition obtain rebuilding pixel, deposit buffer memory in and read for prediction engine B.
Concrete, (wherein each lattice is one 4 * 4 as shown in Figure 7, the numeral of the inside is the label of 4 * 4, also represent the order of processing, represent simultaneously 4 * 4 positions that are arranged in the basic processing unit piece), we take a basic processing unit block size and are example as 32 * 32, introduce the specific works flow process of prediction unit and method in HEVC encoder frames of the present invention.
At first, first complete the scanning of predictive mode and the scanning of piece dividing condition, specific as follows:
Step 1: predict for 4 * 4 to label 0.The control signal of input is followed successively by: the control signal (CU_size) that characterizes the size of current prediction piece is 4, (Blk4 * 4_num) is 0, and the control signal (Mode) that characterizes the pattern of current prediction inputs 0 to 34 successively to characterize the control signal of the position of current processing block in the basic processing unit piece;
Step 2: calculate the SATD value under every kind of pattern, be about to the pixel that prediction obtains poor with corresponding original pixels, then do the Hadamard conversion, get SATD value minimum as optimization model.
Step 3: for label, be 4 * 4 operations of carrying out step 1, step 2 of 1,2,3, control signal Blk4 * 4_num is followed successively by 1,2,3, and two other control signal is constant.
Step 4: the SATD value addition by calculating after prediction under the optimization model of 0,1,2,3 four 4 * 4 obtains by 0,1,2,3 four 4 * 4 optimum SATDs of 8 * 8 of forming in the division situation.
Step 5: predict for 8 * 8 to four 4 * 4 compositions of label 0,1,2,3, the control signal situation of input is: CU_size remains 8, Mode is that 0 o'clock Blk4 * 4_num is followed successively by 0,1,2,3, finish after prediction and the calculating of SATD by the addition of 4 SATD values, as these 8 * 8 SATD under pattern 0.To Mode, be similarly 1 to 34 to do same operation.
Step 6: the SATD of more all patterns, go these 8 * 8 optimization models in not division situation of its minimum conduct.
Step 7: the SATD that comparison step 4 and step 6 obtain, get its smaller as the optimal cases of dividing, if the SATD of step 4 is less, current 8 * 8 block splittings are four 4 * 4; If step 6 must with SATD less, current 8 * 8 are not divided.The final SATD that smaller value in both is the most current 8 * 8.
…………
Step n: deposit optimical block dividing condition and corresponding optimal prediction modes in buffer.
To after each 4 * 4 and 8 * 8 to do the processing of step 1 to step 7, according to the rule of recurrence, 16 * 16 and 32 * 32 operation is also similar, these steps and step 1 are similar to step 7, with the ellipsis replacement, repeat no more.In a word, after handling four 4 * 4, process immediately one 8 * 8 at every turn, after handling four 8 * 8, process immediately one 16 * 16 at every turn, immediately process one 32 * 32 piece after handling four 16 * 16 at every turn, this be namely the present invention adopt postorder traversal.Thereby complete the scanning of predictive mode and the scanning of piece dividing condition.
Steps A 1: first according to control information, read the reference pixel that corresponding original pixels is used as prediction, and its filling of deriving, filtering are processed, the reference pixel that obtains final output is delivered in the general predicting unit of three grades of flowing water.
Steps A 2: to the projection between the reference pixel procession and preliminary selection, namely select the required all reference pixels of current whole prediction piece as standby reference pixel according to control information.In addition, according to control information, calculate in forecasting process and need the parameter iIdx used and iFact.(in formula (2) (3), can find out, in the forecasting process of pattern 2 to 17, for 4 pixels of each row in 4 * 4, parameter iIdx is used in prediction and iFact is same value; Similarly, in the forecasting process of pattern 18 to 34,4 pixels of every delegation, parameter iIdx is used in prediction and iFact is also same value.So this step calculates four iIdx values and four iFact values, processing unit afterwards can be divided into four parallel simple process unit, and each processing unit is processed respectively every delegation or each row of 4 * 4 according to pattern, as shown in Figure 4.）
Steps A 3: from selecting standby reference pixel, the every delegation when 4 * 4 of pre-treatments or each row are predicted to the reference pixel that need to use according to four iIdx values.
Steps A 4: according to iFact and corresponding reference pixel, every delegation or each row pixel when 4 * 4 of pre-treatments are carried out to interpolative prediction, thereby obtain corresponding predict pixel.
Then, when step 1 after the scanning of the scanning traversal of the predictive mode of step n and piece dividing condition has traveled through, the work that enters prediction engine B, its concrete workflow is:
Step B1: read optimical block dividing condition and corresponding optimal prediction modes from buffer, as control information.
Step B2: according to control information, read the reference pixel that corresponding reconstruction pixel is used as prediction, and its filling of deriving, filtering are processed, the reference pixel that obtains final output is delivered in the general predicting unit of three grades of flowing water.
Step B3: with step A2;
Step B4: with step A3;
Step B5: with step A4;
Step B6: according to final predict pixel and corresponding original pixels, calculate residual error;
Step B7: wait for completing of process of reconstruction, to start processing procedure next time.
After the work of prediction engine B completes, namely enter the process of rebuilding loop, namely residual error carried out conversion (DCT/DST), quantized, obtain the residual error coefficient of final output, then pass through inverse quantization, inverse transformation (IDCT/IDST) obtains rebuilding pixel, deposits buffer memory in.
In the embodiment of this programme, the judgement selection course that the prediction of the scanning of basic processing unit predictive mode and piece dividing condition traversal and pattern and piece are divided is placed in the first order of system flow waterline, prediction and the computational process of residual error and the second level that the reconstruction loop is placed in the system flow waterline of carrying out according to optimization model and optimical block dividing condition.The forecasting process of the first order be take original pixels and is reference pixel, so the process of reconstruction that the prediction that next processing block is carried out does not need to wait for current block finishes to carry out, reduced the dependence of data, the basic processing unit piece of one 32 * 32 is finished to the traversal of all situations and need to be done 8960 predictions (each 16 pixels of a prediction processing i.e. block sizes are 4 * 4), between each prediction, do not have the idle running of clock; The forecasting process of the second level be take and rebuild pixel and be reference pixel, so need to waiting for after the process of reconstruction of current block finishes, the prediction that a processing block is carried out just can carry out, but need not carry out the traversal of all situations, the basic processing unit piece of handling one 32 * 32 only need to be done 64 predictions, between every twice prediction, there is the idle running of clock, to wait for completing of rebuilding.The time of two-stage can well mate, when the second level during in the processing of basic processing unit piece of carrying out one 32 * 32, the first order can be processed next 32 * 32 basic processing unit piece, thereby forms effective pipeline organization like this, as shown in Figure 3, improved throughput.
In the embodiment of this programme, the scanning traversal of basic processing unit piece dividing condition, adopt the mode of postorder traversal, as shown in Figure 5, namely from " leaf " (4 * 4 size block) of lowermost layer, start to operate, handle four " leaves " operates their " root " at every turn, continue by that analogy, each like this handle four " leaves " and one " root ", just can from two kinds, select one as best dividing condition, another thoroughly to be abandoned, need not store, thereby save the area of intermediate variable memory.
In the embodiment of this programme, the scanning of basic processing unit block prediction mode traversal, therefore the unit of processing due to fallout predictor is namely 4 * 4, for prediction block sizes, is 4 * 4, the prediction of carrying out successively 35 kinds of patterns gets final product.And, for larger piece, while carrying out a kind of prediction of pattern, all to handle successively its each 4 * 4 sub-block at every turn, then carry out the prediction of lower a kind of pattern, as shown in Figure 6.For bulk, every kind of pattern is finished like this, and the storage (keeping initial value still to replace with the pattern just completed) that only need update optimization model gets final product, and has also saved to a certain extent the size of memory.
Claims (9)
1. prediction unit in an encoder frames that is applicable to the HEVC standard is characterized in that comprising:
A control unit, while in order to be controlled at, encoding, 35 kinds of traversal order that predictive mode is predicted respectively that traversal order and the employing standard of the quad-tree partition situation of basic processing unit piece proposed, in addition also in order to control in some forecasting processes the stand-by period to process of reconstruction, to meet the sequential requirement;
Two prediction engine: prediction engine A and prediction engine B, wherein: prediction engine A: the original pixels of take is reference pixel, current basic processing unit piece is predicted, to all possible dividing condition, all adopt 35 kinds of predictive modes to predict, then compare to determine and select optimization model and best dividing condition; Prediction engine B: the reconstruction pixel of take is reference pixel, and current basic processing unit piece is predicted, only for optimization model and best dividing condition, predicts, then with original pixels, makes the poor residual error that obtains;
Rebuild loop for one, carry out conversion, quantification in order to the residual values to the basic processing unit piece, the residual error coefficient that obtains exporting, then obtain rebuilding pixel through inverse quantization and inverse transformation;
An original pixels buffer, in order to the original pixel value when pretreatment unit piece and the peripheral piece that need to use that storage is read, A provides reference pixel for prediction engine, and provides for prediction engine B the input of asking residual error;
Rebuild the picture element caching device for one, rebuild in order to storage the reconstruction pixel that loop obtains, B provides reference pixel for prediction engine;
Pattern, piece are divided the result cache device, the optimal prediction modes obtained after the traverse scanning of prediction engine A in order to storage and the result of optimum division situation, and B reads for prediction engine, to control the process of its prediction.
2. prediction unit in encoder frames according to claim 1, is characterized in that, described control unit comprises a state machine and a plurality of counter:
Described state machine, be used to controlling 4 * 4,8 * 8,16 * 16,32 * 32 forecasting process, the method scanning quad-tree structure of employing postorder traversal;
Described counter, for controlling the forecasting process of different masses size, the order of mode scanning and the order of processing fritter in bulk.
3. prediction unit in encoder frames according to claim 1, is characterized in that, described prediction engine A comprises a reference pixel processing unit, a general predicting unit and a judgement selected cell; Described prediction engine B comprises a reference pixel processing unit, a general predicting unit and a residual computations unit; Wherein:
Described reference pixel processing unit, process in order to filling that reference pixel is derived, filtering; It comprises 129 registers, in order to the stored reference pixel, uses 17 when carrying out 4 * 4 predictions, uses 33 while carrying out 8 * 8 predictions, uses 65 while carrying out 16 * 16 predictions, uses 129 while carrying out 32 * 32 predictions;
Described general predicting unit, be three grades of fully pipelined architectures of 16 pixel parallel processings, can support the forecasting process of all 35 kinds of patterns of 4 * 4,8 * 8,16 * 16,32 * 32 block sizes;
Described judgement selected cell, calculate the SATD value when 4 * 4 of pre-treatments, namely by predicted value and original value, subtract each other, difference is done to the Hadamard conversion, relatively choose the predictive mode of the conduct optimum that the SATD value is less and best piece division by corresponding cumulative sum;
Described residual computations unit, subtract each other with the pixel that prediction obtains the residual values obtained be used to calculating original pixels.
4. prediction unit in encoder frames according to claim 3, it is characterized in that, described three grades of fully pipelined architectures have three control signals: characterize " Mode " signal of the predictive mode of current employing, characterize " CU_size " signal of current prediction block sizes and characterize 4 * 4 " Blk4 * 4_num " signals that are arranged in the position of basic processing unit piece when pre-treatment; In described three grades of fully pipelined architectures,
(1) the streamline first order comprises a reference pixel projection and initial option unit, and reference pixel is carried out to preliminary locating
The selection of reason; Coordinate computing unit and the Prediction Parameters computing unit of one current 4 * 4, in order to calculate according to control signal the parameter that forecasting process need to be used; A pre-measuring angle look-up table, in order to therefrom to search angle value to calculate relevant parameter according to current predictive mode;
(2) second level of streamline comprises four reference pixel selectors, according to the parameter that previous stage is calculated, right respectively
When each row or each row of 4 * 4 of pre-treatments, select corresponding reference pixel;
(3) third level of streamline comprises four interpolation devices, in order to the reference pixel that utilizes previous stage to obtain, respectively to current
Process each row of 4 * 4 or respectively be listed as the process of carrying out interpolative prediction.
5. an encoder intra-frame prediction method that is applicable to the HEVC standard, is characterized in that, concrete steps are as follows:
(1) take original pixels is reference pixel, at first carries out the predictive mode of basic processing unit piece and the scanning traversal forecasting process of piece dividing condition, and therefrom selects optimum predictive mode and best piece dividing condition;
(2) take and rebuild pixel and be reference pixel, carry out again the forecasting process according to optimal prediction modes and optimical block dividing condition, obtain final predicted value, calculate afterwards residual error, and it is carried out to DCT or DST conversion, quantification, the residual error coefficient that obtains exporting, then carry out inverse quantization, IDCT or IDST inverse transformation, obtain rebuilding pixel value.
6. encoder intra-frame prediction method according to claim 5, it is characterized in that: in step 1), select optimum predictive mode and best piece dividing condition, the SATD value that the Hadamard conversion of calculating 4 * 4 of take obtains is benchmark, the SATD value sum of four 4 * 4 is the SATD value of 8 * 8, in like manner obtain 16 * 16 and 32 * 32 SATD value, SATD value smaller is optimal situation.
7. encoder intra-frame prediction method according to claim 5, is characterized in that, in described scanning traversal forecasting process, for the scanning that the block structure of quaternary tree is divided, employing be the method for postorder traversal, concrete step is:
(1) from the upper left side of basic processing unit, first to first 4 * 4, adopt 35 kinds of predictive modes to predict,
According to the order of Z scanning, successively its right, 35 kinds of predictive modes of three 4 * 4 the employings following and limit, bottom right are predicted again;
(2) after finishing the prediction of four 4 * 4,8 * 8 of this four 4 * 4 compositions are carried out to the prediction that prediction block sizes is 35 kinds of predictive modes of 8 * 8;
(3) after finishing this prediction of 8 * 8, according to the order of Z scanning, successively to its right, following and the limit, bottom right
Three 8 * 8 processing procedures of carrying out step (1), (2);
(4), after finishing the prediction of four 8 * 8,16 * 16 of this four 8 * 8 compositions are carried out to prediction block sizes be
The prediction of 35 kinds of predictive modes of 16 * 16;
(5) after finishing this prediction of 16 * 16, according to the order of Z scanning, successively to its right, following and limit, bottom right
Three 16 * 16 processing procedures of carrying out step (1) (2) (3) (4);
(6) after finishing the prediction of four 16 * 16,32 * 32 of this four 16 * 16 compositions are predicted to piece is large
Little is the prediction of 35 kinds of predictive modes of 32 * 32.
8. encoder intra-frame prediction method according to claim 5, is characterized in that, in described scanning traversal forecasting process, 35 kinds of predictive modes scanned, and concrete step is:
(1) for the scanning of 4 * 4, because the parallel processing element of forecasting process is 16 pixels, namely one 4 * 4,
While like this pattern being scanned, the prediction that only need to carry out successively 35 kinds of patterns gets final product;
(2) for the scanning of bulk, take and 8 * 8 be example, first with a kind of predictive mode, its four 4 * 4 of comprising are predicted respectively, and then change another kind of pattern and equally four 4 * 4 are predicted, until traveled through all patterns; 16 * 16 and 32 * 32 adopt similar method.
9. encoder intra-frame prediction method according to claim 5, is characterized in that, in step (1) and step (2), before forecasting process, all derive filling, filtering processed to reference pixel, and process comprises:
(1) which reference pixel of judgement periphery exists, and which reference pixel does not exist;
(2) utilize the reference pixel existed to derive to non-existent reference pixel, be filled into the position of non-existent reference pixel;
(3) reference pixel is carried out to the filtering processing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013103216770A CN103414895A (en) | 2013-07-29 | 2013-07-29 | Encoder intra-frame prediction device and method applicable to HEVC standards |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013103216770A CN103414895A (en) | 2013-07-29 | 2013-07-29 | Encoder intra-frame prediction device and method applicable to HEVC standards |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103414895A true CN103414895A (en) | 2013-11-27 |
Family
ID=49607877
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013103216770A Pending CN103414895A (en) | 2013-07-29 | 2013-07-29 | Encoder intra-frame prediction device and method applicable to HEVC standards |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103414895A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103888782A (en) * | 2014-03-04 | 2014-06-25 | 上海交通大学 | Parallel task partitioning method for HEVC decoder |
| CN104363458A (en) * | 2014-10-29 | 2015-02-18 | 复旦大学 | Hardware addressing method for prediction units used for intra-frame prediction in HEVC (high efficiency video coding) standard |
| 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 |
| CN105072444A (en) * | 2015-07-31 | 2015-11-18 | 宁波大学 | Detection method for HEVC video secondary compression with different quantization parameters |
| CN105100804A (en) * | 2014-05-20 | 2015-11-25 | 炬芯（珠海）科技有限公司 | Method and device for video decoding |
| 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 |
| WO2016091122A1 (en) * | 2014-12-11 | 2016-06-16 | Mediatek Inc. | Method and apparatus of intra prediction in video coding |
| CN106911935A (en) * | 2017-04-19 | 2017-06-30 | 西安电子科技大学 | Method of designing integrated circuit based on HEVC entropy coders |
| CN107071474A (en) * | 2017-04-13 | 2017-08-18 | 西安电子科技大学 | The HEVC intra-frame prediction methods being multiplexed based on SATD |
| CN107079167A (en) * | 2014-10-01 | 2017-08-18 | 高通股份有限公司 | Pipelining infra-frame prediction hardware structure for video coding |
| CN107682707A (en) * | 2017-10-23 | 2018-02-09 | 福州大学 | Meet the image dividing system based on quad-tree structure of H.265 standard |
| CN110868596A (en) * | 2019-11-21 | 2020-03-06 | 北京数码视讯科技股份有限公司 | Distributed encoding method and device, electronic equipment and readable storage medium |
| WO2020063598A1 (en) * | 2018-09-25 | 2020-04-02 | Huawei Technologies Co., Ltd. | A video encoder, a video decoder and corresponding methods |
| WO2020119525A1 (en) * | 2018-12-13 | 2020-06-18 | 华为技术有限公司 | Inter-frame prediction method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101820546A (en) * | 2009-02-27 | 2010-09-01 | 源见科技（苏州）有限公司 | Intra-frame prediction method |
| US20110274158A1 (en) * | 2010-05-10 | 2011-11-10 | Mediatek Inc. | Method and Apparatus of Adaptive Loop Filtering |
| US20120082210A1 (en) * | 2010-10-01 | 2012-04-05 | Qualcomm Incorporated | Coding prediction modes in video coding |
| CN102843559A (en) * | 2012-09-12 | 2012-12-26 | 清华大学 | Method and device for quickly selecting HEVC intra prediction mode on basis of texture characteristics |
| CN103024392A (en) * | 2012-12-26 | 2013-04-03 | 清华大学 | Method and device for intra-frame mode prediction based on two-dimensional Hadamard transformation |
-
2013
- 2013-07-29 CN CN2013103216770A patent/CN103414895A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101820546A (en) * | 2009-02-27 | 2010-09-01 | 源见科技（苏州）有限公司 | Intra-frame prediction method |
| US20110274158A1 (en) * | 2010-05-10 | 2011-11-10 | Mediatek Inc. | Method and Apparatus of Adaptive Loop Filtering |
| US20120082210A1 (en) * | 2010-10-01 | 2012-04-05 | Qualcomm Incorporated | Coding prediction modes in video coding |
| CN102843559A (en) * | 2012-09-12 | 2012-12-26 | 清华大学 | Method and device for quickly selecting HEVC intra prediction mode on basis of texture characteristics |
| CN103024392A (en) * | 2012-12-26 | 2013-04-03 | 清华大学 | Method and device for intra-frame mode prediction based on two-dimensional Hadamard transformation |
Non-Patent Citations (1)
| Title |
|---|
| 严顺卿: "HEVC帧内预测快速模式选择算法研究", 《上海交通大学硕士学位论文》 * |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103888782B (en) * | 2014-03-04 | 2017-02-08 | 上海交通大学 | Parallel task partitioning method for HEVC decoder |
| CN103888782A (en) * | 2014-03-04 | 2014-06-25 | 上海交通大学 | Parallel task partitioning method for HEVC decoder |
| CN105100804A (en) * | 2014-05-20 | 2015-11-25 | 炬芯（珠海）科技有限公司 | Method and device for video decoding |
| CN107079167A (en) * | 2014-10-01 | 2017-08-18 | 高通股份有限公司 | Pipelining infra-frame prediction hardware structure for video coding |
| CN107079167B (en) * | 2014-10-01 | 2020-07-17 | 高通股份有限公司 | Method, apparatus and system for performing intra prediction |
| CN104363458A (en) * | 2014-10-29 | 2015-02-18 | 复旦大学 | Hardware addressing method for prediction units used for intra-frame prediction in HEVC (high efficiency video coding) standard |
| CN104363458B (en) * | 2014-10-29 | 2017-10-10 | 复旦大学 | A kind of hardware addressing addressing method for being used for the predicting unit of infra-frame prediction in HEVC standard |
| US10412402B2 (en) | 2014-12-11 | 2019-09-10 | Mediatek Inc. | Method and apparatus of intra prediction in video coding |
| WO2016091122A1 (en) * | 2014-12-11 | 2016-06-16 | Mediatek Inc. | Method and apparatus of intra prediction in video coding |
| 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 |
| CN104811696B (en) * | 2015-04-17 | 2018-01-02 | 北京奇艺世纪科技有限公司 | A kind of coding method of video data and device |
| CN105072444B (en) * | 2015-07-31 | 2018-01-05 | 宁波大学 | A kind of HEVC video second-compressed detection methods under different quantization parameters |
| CN105072444A (en) * | 2015-07-31 | 2015-11-18 | 宁波大学 | Detection method for HEVC video secondary compression with different quantization parameters |
| 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 |
| CN105516728B (en) * | 2015-12-15 | 2019-06-28 | 华中科技大学 | A kind of parallel intra-frame prediction method of H.265/HEVC middle 8x8 sub-macroblock |
| CN107071474A (en) * | 2017-04-13 | 2017-08-18 | 西安电子科技大学 | The HEVC intra-frame prediction methods being multiplexed based on SATD |
| CN106911935A (en) * | 2017-04-19 | 2017-06-30 | 西安电子科技大学 | Method of designing integrated circuit based on HEVC entropy coders |
| CN107682707A (en) * | 2017-10-23 | 2018-02-09 | 福州大学 | Meet the image dividing system based on quad-tree structure of H.265 standard |
| CN107682707B (en) * | 2017-10-23 | 2023-06-09 | 福州大学 | Image dividing system conforming to H.265 standard and based on quadtree structure |
| WO2020063598A1 (en) * | 2018-09-25 | 2020-04-02 | Huawei Technologies Co., Ltd. | A video encoder, a video decoder and corresponding methods |
| WO2020119525A1 (en) * | 2018-12-13 | 2020-06-18 | 华为技术有限公司 | Inter-frame prediction method and device |
| CN110868596A (en) * | 2019-11-21 | 2020-03-06 | 北京数码视讯科技股份有限公司 | Distributed encoding method and device, electronic equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103414895A (en) | Encoder intra-frame prediction device and method applicable to HEVC standards | |
| CN102547296B (en) | Motion estimation accelerating circuit and motion estimation method as well as loop filtering accelerating circuit | |
| CN104244007B (en) | Image coding method and device and decoding method and device | |
| CN101361370B (en) | Image encoding/image decoding method and image encoding/image decoding apparatus | |
| CN104883578B (en) | The method decoded to intra prediction mode | |
| CN102065298B (en) | High-performance macroblock coding implementation method | |
| CN103248895B (en) | A kind of quick mode method of estimation for HEVC intraframe coding | |
| CN105791829B (en) | A kind of parallel intra-frame prediction method of HEVC based on multi-core platform | |
| CN104702958A (en) | HEVC intraframe coding method and system based on spatial correlation | |
| CN111314698A (en) | Image coding processing method and device | |
| CN109068142A (en) | 360 degree of video intra-frame prediction high-speed decisions based on textural characteristics | |
| CN109495743A (en) | A kind of parallelization method for video coding based on isomery many places platform | |
| CN101505427A (en) | Movement estimation apparatus in video compression encoding algorithm | |
| CN104754338A (en) | Selection method and device for intra-frame predication mode | |
| CN1589028B (en) | Predicting device and method based on pixel flowing frame | |
| CN101115207B (en) | Method and device for implementing interframe forecast based on relativity between future positions | |
| CN102291581A (en) | Realizing method of self-adaptive motion estimation supporting frame field | |
| CN107087171A (en) | HEVC integer pixel motion estimation methods and device | |
| CN102769753B (en) | H264 encoder and coding method | |
| CN110419214A (en) | Intra prediction mode searching method and device, method for video coding and device and recording medium | |
| CN110198442B (en) | Intra-frame prediction method, device and storage medium for video coding | |
| CN110365988A (en) | A kind of H.265 coding method and device | |
| Liu et al. | A highly pipelined VLSI architecture for all modes and block sizes intra prediction in HEVC encoder | |
| CN105100799B (en) | A method of reducing intraframe coding time delay in HEVC encoders | |
| CN102801982B (en) | Estimation method applied on video compression and based on quick movement of block integration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20131127 |
|
| WD01 | Invention patent application deemed withdrawn after publication |