CN103702131B - Pattern-preprocessing-based intraframe coding optimization method and system - Google Patents

Abstract

The invention provides a pattern-preprocessing-based intraframe coding optimization method and a pattern-preprocessing-based intraframe coding optimization system. The method comprises the following steps of 1, dividing a coding tree unit of an original image into a plurality of N*N coding units, and calculating the edge strength of each point in the coding units, wherein N belongs to {4,8,16,32}; 2, classifying all the N*N coding units based on coding unit edge strength analysis, wherein N belongs to {4,8,16,32}; 3, selecting a corresponding residual estimation model according to a classification result, and performing pixel prediction error estimation on each N*N coding unit, wherein N belongs to {4,8,16,32}; 4, calculating the overall coding rate distortion cost RDN of each N*N coding unit according to the pixel prediction error estimation of the N*N coding unit, wherein N belongs to {4,8,16,32}; 5, when N belongs to {4,32}, calculating the segmented coding cost of each N*N coding unit, namely cost in the segmentation of each N*N coding unit into four corresponding coding units; 6, at the moment, comparing an overall coding rate distortion cost value and a segmented coding cost value of each N*N coding unit, not segmenting the N*N coding unit if the overall coding rate distortion cost value is less than the segmented coding cost value, otherwise segmenting the N*N coding unit into four corresponding coding units.

Description

A kind of intraframe coding optimization method based on pattern pretreatment and system

Technical field

The present invention relates to technical field of video processing is and in particular to a kind of intraframe coding optimization side based on pattern pretreatment Method and system.

Background technology

High efficiency Video coding（HEVC, High Efficiency Video Coding）Be continue H.264/AVC standard it More popular coding standard afterwards.In order to deal carefully with high definition, the image sequence under ultra high-definition resolution ratio, HEVC proposes a kind of multiple Close coding unit（CTU, coding tree unit）.The complete image of one frame, no matter its resolution sizes, start in its coding When, all can be divided into the code tree unit being made up of 64 × 64 luminance pixel dot matrix and corresponding chroma pixel dot matrix CTU, hereafter again in units of CTU, is encoded successively.Because chroma coder is similar to brightness, complexity is simple compared with brightness, because This, with regard to the discussion of HEVC encoder complexity, typically concentrate on the coding of its luminance component.CTU in an encoding process, further It is divided into encoding block (CU, coding unit）, prediction block (PU, prediction unit) and transform block (TU, transform unit).In HEVC, the segmentation of CTU is in coding, prediction and conversion stages, according to its meaning, using different Segmentation, forms the segmentation combination of CU/PU/TU.The concrete division of CU, all employs the strategy of quaternary tree Layering memory.CU is split The root node of quaternary tree is maximum encoding block, in the same size with CTU, usually 64 × 64.And segmentation the most tiny, then by 8 × 8 Pixel-matrix forms, i.e. minimum code block.Between maximum encoding block and minimum code block, then by the way of layering segmentation, Every layer of CU by 2N × 2N size is divided into the CU of 4 N × N sizes.The size of PU is identical with corresponding CU, and its content is because predicting mould Formula and different.Predictive mode one has 35, direct current（DC,Direct Circuit）Prediction, plane（planer）Prediction, and 33 Direction prediction.Forecast period, is choose optimum from 35 predictive modes one, so that it is determined that PU, carries out next step volume Code.The division of TU divides identical strategy using with CU, and its effect is in conversion stages, a certain CU/PU mode combinations to be determined Residual error carry out optimal transformation, that is, for the residual matrix determining, find optimal partitioning scheme, at utmost to pass through to convert Remove information redundancy so that conversion after required coding information content minimum.The combination of CU/PU/TU three, constitutes candidate's mould Formula vector, withRepresent it.In HEVC coding, the search procedure of model prediction, is to find minimum code cost for principle time Go through all combinations.Under candidate pattern vectorCoding cost, it is represented with J, be with image domains pass through distortion rate cost (RD-Cost, Rate Distortion Cost) function, to obtain, is shown below：

Wherein,It is in pattern vectorUnder, the code check that obtained by entropy coding,It is in pattern vector Under coding distortion rate, obtained by the difference of reconstructed value and the original value of picture element matrix, λ is Lagrange factor, for controllingWithShared weight.

WithAccurate calculating, inevitable must need execution coding and process of reconstruction.Cataloged procedure mainly wraps Include block prediction, residual energy calculates, two-dimensional transform quantifies；And rebuild main inclusion inverse quantization, and two-dimentional inverse transformation, reconstructed value meter Calculate etc..Wherein residual energy PE_kIt is encoding block pixel value C_kWith prediction block pixel value P_kDifference square, that is,

PE_k=(P_k-C_k)²,

Wherein, k=i × N+j, represents the i-th row in N × N block, the pixel value of jth row.And the quantization of quantizing process institute foundation Step-length Qs is only relevant with coder parameters Qp, is shown below：

Wherein,Represent downward floor operation, % represents modulo operation.

Encoder must expend huge time and power consumption, executes conversion, quantization, inverse quantization, inverse transformation, entropy code waited Journey, thus obtainWithExact value.And pattern vectorErgodic process, allow complexity double.Multiplication factor with The mode combinations number of CU/PU/TU is relevant.And the mode combinations number overwhelming majority of CU/PU/TU is depending on the number of CU, great Liang Shi Test and show, encoder complexity acutely increases with the increase of CU quaternary tree depth.

In sum, the inner frame coding method based on CU/PU/TU pattern, is a kind of exhaust algorithm, encoder complexity is relatively Height is it is therefore desirable to be optimized to based on the inner frame coding method of CU/PU/TU pattern, to reduce encoder complexity.It is currently based on The intraframe coding optimization method of pattern is broadly divided into three kinds：The first simplifies the calculation of RD-Cost；It two is to filter maximum Can not possibly candidate pattern；Its three be setting Coding cost threshold value to realize terminating in advance.

Three of the above optimization method effectively reduces the amount of calculation of intraframe predictive coding, almost can prop up on average Hold HEVC real-time coding.But, these method complexities are unstable, and when continuously complex situations, coding rate is relatively low, because This can not fundamentally ensure the realization of real-time coding function.

Content of the invention

For the deficiencies in the prior art, the present invention provides the intraframe coding optimization method and system based on pattern pretreatment, So that can stably accelerate coding rate in HEVC encoding-operation process.

For achieving the above object, the present invention is achieved by the following technical programs：

A kind of intraframe coding optimization method based on pattern pretreatment, the method includes：

S1, the code tree unit of original image is divided into the encoding block of several N × N, N ∈ { 4,8,16,32 }, calculate and compile The edge strength of every bit in code block；

S2, the encoding block to all N × N carry out the classification based on the analysis of encoding block edge strength, N ∈ { 4,8,16,32 }；

S3, corresponding residual error is selected to estimate that model carries out pixel predicated error to each N × N encoding block according to classification results Estimate, N ∈ { 4,8,16,32 }；

S4, the pixel predicated error estimation according to each N × N encoding block, calculate the binary encoding generation of each N × N encoding block Valency RD_N, N ∈ { 4,8,16,32 }；

S5, when N ∈ { 8,32 }, calculate each N × N code block segmentation Coding cost, that is, be divided into 4 corresponding Cost required for encoding block, is designated as

S6, when N ∈ { 8,32 }, the RD of relatively each N × N encoding block_NWithValue, ifThen this N × N Encoding block is not split, otherwise, this N × N code block segmentation is corresponding for 4Encoding block.

Wherein, described step S1 includes：The code tree unit of original image is divided into the encoding block of several N × N, N ∈ { 4,8,16,32 }, the edge strength calculating (i, j) position in the encoding block of N × N is ES_k,

Wherein, k=i × N+j, eh_kFor its horizontal component, ev_kFor its vertical component, work as m, during n ∈ { 0,1 ..., N-1 }, C_m,nRepresent the pixel value at position (m, n) place in current N × N encoding block, as m ∈ { -1, N } or n ∈ { -1, N }, C_m,nRepresent it Neighborhood territory pixel value.

Wherein, described step S2 includes：

S21, the encoding block for each N × N, the edge strength of its all pixels point are projected paramount according to its edge direction 33 PU model prediction directions in efficiency video coding HEVC, find out the maximum side of each N × N encoding block edge strength projection To as the principal direction of this encoding block；

S22, foundation encoding block principal direction classification, totally 4 class：

According to encoding block principal direction and the level interval D pre-setting₀:[7,14), vertically interval D₁:[23,30), minus 45 The interval D of degree₂:[14,23), other interval D₃, judge which interval encoding block falls in；

S23, the homogeney of foundation encoding block edge strength projection are classified, totally 2 class：

Calculation code block principal direction and its neighbouring four direction edge strength projection sum, are designated as σ；Calculate this encoding block The edge strength projection sum in all 33 directions, is designated as Σ；Judge whether following formula is set up：

σ/Σ ＞ 1-0.1 × log₂N,

If above formula is true, the edge strength projection of this encoding block is referred to as homogeneity, otherwise referred to as heterogeneous；

S24, classified according to pixel edge strength maximum in encoding block, totally 7 class：

Find out the maximum of pixel edge strength in encoding blockAccording to set in advance The class interval { 0,400,900,1600,2500,4900,8100, ∞ } of class separation composition, judges which encoding block belongs to Class, wherein, the adjacent interval folded by class separation of each two is class interval, totally seven class intervals；

S25, step S22, S23 and S24 constitute altogether 56 classes, to each encoding block, successively according to step S21, S22, S23, S24 find corresponding classification sequence number in 56 classes.

Wherein, described step S3 includes：

If the classification sequence of a certain N × N encoding block is l, quantization step is Qs, and pixel edge strength is ES_k, calculating should The pixel predicated error estimate of encoding block

Wherein, a (l, N), b_k(l,N),k∈{0,1,...,N²- 1 } for the ginseng of l class N × N encoding block residual prediction model Number, l ∈ { 0,1 ..., 55 }.

Wherein, described step S4 includes：

When N ∈ { 4,8,16,32 }, binary encoding cost RD of each N × N encoding block_NComputing formula is as follows：

Wherein, w_rFor bit rate weight parameter, its value withTri- amounts of Qs, N are relevant, when encoder is realized to search The mode of table is realized, w_dFor distortion rate weight parameter, it is shown below：

Wherein, described step S5 includes：

During N ∈ { 8,32 }, calculate each N × N code block segmentation Coding cost, that is, be divided into 4 correspondingCoding Cost required for block is designated as

Wherein,Represent 4 that N × N code block segmentation is becomeEncoding block binary encoding Cost is sued for peace；γ_mode=4, represent the introduced cost of bit that partition encoding increases, γ_cbf=1, it is to represent coded block mark Introduced cost.

A kind of intraframe coding based on pattern pretreatment optimizes system, and this system includes：

Edge strength computing module, for being divided into the encoding block of several N × N, N ∈ by the code tree unit of original image { 4,8,16,32 }, the edge strength of every bit in calculation code block；

Encoding block sort module, for the encoding block of all N × N is carried out based on the analysis of encoding block edge strength point Class, N ∈ { 4,8,16,32 }；

Block prediction estimation error module, for selecting corresponding residual error to estimate model to each N × N according to classification results Encoding block carries out pixel predicated error estimation, N ∈ { 4,8,16,32 }；

Binary encoding cost computing module, estimates for the pixel predicated error according to each N × N encoding block, calculates each N Binary encoding cost RD of × N encoding block_N, N ∈ { 4,8,16,32 }；

Partition encoding cost computing module, for calculating each N × N code block segmentation Coding cost, that is, is divided into 4 correspondences 'sCost required for encoding block, is designated as

Whether split determination module, for when N ∈ { 8,32 }, the RD of relatively each N × N encoding block_NWithValue, ifThen this N × N encoding block is not split, otherwise, this N × N code block segmentation is corresponding for 4Coding Block.

Wherein, described edge strength computing module is used for the code tree unit of original image is divided into the volume of several N × N Code block, N ∈ { 4,8,16,32 }, the edge strength calculating (i, j) position in the encoding block of N × N is ES_k：

Wherein, k=i × N+j, eh_kFor its horizontal component, ev_kFor its vertical component, work as m, during n ∈ { 0,1 ..., N-1 }, C_m,nRepresent the pixel value at position (m, n) place in current N × N encoding block, as m ∈ { -1, N } or n ∈ { -1, N }, C_m,nRepresent it Neighborhood territory pixel value.

Wherein, described encoding block sort module includes：

Projection subelement, for the encoding block for each N × N, by the edge strength of its all pixels point according to its edge Direction projection, to 33 PU model prediction directions in high efficiency video coding HEVC, is found out each N × N encoding block edge strength and is thrown The maximum direction of shadow, as the principal direction of this encoding block；

First classification subelement, for according to the classification of encoding block principal direction, totally 4 class：

According to encoding block principal direction and the level interval D pre-setting₀:[7,14), vertically interval D₁:[23,30), minus 45 The interval D of degree₂:[14,23), other interval D₃, judge which interval encoding block falls in；

Second classification subelement, for being classified according to the homogeney of encoding block edge strength projection, totally 2 class：

Calculation code block principal direction and its neighbouring four direction edge strength projection sum, are designated as σ；Calculate this encoding block The edge strength projection sum in all 33 directions, is designated as Σ；Judge whether following formula is set up：

σ/Σ ＞ 1-0.1 × log₂N,

If above formula is true, the edge strength projection of this encoding block is referred to as homogeneity, otherwise referred to as heterogeneous；

3rd classification subelement, for being classified according to pixel edge strength maximum in encoding block, totally 7 class：

Find out the maximum of pixel edge strength in encoding blockAccording to set in advance The class interval { 0,400,900,1600,2500,4900,8100, ∞ } of class separation composition, judges which encoding block belongs to Class, wherein, the adjacent interval folded by class separation of each two is class interval, totally seven class intervals；

Classification subelement, for each encoding block, successively according to projection subelement, the first classification subelement, second Classification subelement and the 3rd classification subelement are from the first classification subelement, the second classification subelement and the 3rd classification subelement composition 56 kinds of classification results in find corresponding classification sequence number.

Wherein, described block prediction estimation error module is used for when the classification sequence of a certain N × N encoding block is l, quantifies Step-length is Qs, and pixel edge strength is ES_kWhen, calculate pixel predicated error estimate

Wherein, a (l, N), b_k(l,N),k∈{0,1,...,N²- 1 } for the ginseng of l class N × N encoding block residual prediction model Number, l ∈ { 0,1 ..., 55 }.

Wherein, described binary encoding cost computing module is used for, when N ∈ { 4,8,16,32 }, calculating each N × N encoding block Binary encoding cost RD_N：

Wherein, w_rFor bit rate weight parameter, its value withTri- amounts of Qs, N are relevant, when encoder is realized to search The mode of table is realized, w_dFor distortion rate weight parameter, it is shown below：

Wherein, described partition encoding cost computing module is used for, when N ∈ { 8,32 }, calculating each N × N code block segmentation and compiling Code cost, that is, be divided into 4 correspondingCost required for encoding block is designated as

Wherein,Represent 4 that N × N code block segmentation is becomeEncoding block binary encoding Cost is sued for peace；γ_mode=4, represent the introduced cost of bit that partition encoding increases, γ_cbf=1, it is to represent coded block mark Introduced cost.

The present invention has following beneficial effect：

The code tree unit CTU of original image is divided into the encoding block CU of several N × N by the present invention first, N ∈ 4, 8,16,32 }, in calculation code block CU every bit edge strength, then each encoding block CU is classified according to edge strength, Each encoding block is selected suitable residual error estimate that model to calculate the whole of each 32 × 32,8 × 8 encoding block CU according to classification results Body Coding cost and code block segmentation cost, thus judging each 32 × 32 further, whether 8 × 8 encoding blocks should be divided, if 32 × 32,8 × 8 encoding blocks all should be split, then coding work below only need to pay close attention to 16 × 16 and 4 × 4 encoding blocks, if only 32 × One of 32 or 8 × 8 encoding blocks should be split, then only need to pay close attention to 16 × 16 and 8 × 8 encoding blocks, or 32 × 32 and 4 × 4 volumes Code block is so that the number of encoding block CU reduces half, thus the mode combinations number of CU/PU/TU reduces so that coding is complicated Degree reduces, and chooses a kind of in the encoding block additionally, due to last selection 32 × 32,16 × 16, and 8 × 8, in 4 × 4 encoding block One kind, thus being suitable to two groups of pattern search engine Parallel Implementation, thus stably accelerating cataloged procedure.Further, since The number of encoding block CU, before CTU coding, is first reduced by method proposed by the present invention, thus without causing coding delay, from And real-time coding can be realized.

Brief description

In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing Have technology description in required use accompanying drawing be briefly described it should be apparent that, drawings in the following description are the present invention Some embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis These accompanying drawings obtain other accompanying drawings.

Fig. 1 is the flow chart of the intraframe coding optimization method in the embodiment of the present invention 1 based on pattern pretreatment；

Fig. 2 is the structural representation optimizing system in the embodiment of the present invention 2 based on the intraframe coding of pattern pretreatment.

Specific embodiment

Purpose, technical scheme and advantage for making the embodiment of the present invention are clearer, below in conjunction with the embodiment of the present invention In accompanying drawing, the technical scheme in the embodiment of the present invention is carried out with clear, complete description it is clear that described embodiment is The a part of embodiment of the present invention, rather than whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art The every other embodiment being obtained under the premise of not making creative work, broadly falls into the scope of protection of the invention.

Embodiment 1

The embodiment of the present invention 1 proposes a kind of intraframe coding optimization method based on pattern pretreatment, and the method is applied to Before code tree unit CTU coding, code optimization process is carried out to the pixel-matrix of code tree unit CTU, specially from 32/16 Choose a kind of in two kinds of encoding block CU, choose result and be referred to as LCB（Large Code Block）, from 8/4 two kinds of encoding block CU Choose a kind of, choose result and be referred to as SCB (Small Code Block) so that intraframe coding amount of calculation is reduced close to one Half, and can be with two groups of pattern search engine implementation parallel encodings, thus stably accelerating cataloged procedure.Referring to Fig. 1, described Method includes step：

Step 101：The code tree unit of original image is divided into the encoding block of several N × N, N ∈ { 4,8,16,32 }, The edge strength of every bit in calculation code block.

In this step, we give tacit consent to the code tree unit CTU of original image and are greater than 32 × 32, during coding, first Code tree unit CTU is divided into the encoding block CU of several N × N, N ∈ { 4,8,16,32 }, with the application under HEVC standard is Its 64 × 64 code tree unit, during coding, is divided into 4 32 × 32 encoding blocks by example first, then by each 32 × 32 coding Block is divided into the encoding block of 4 16 × 16, and the rest may be inferred, till being partitioned into 4 × 4 encoding block, then carries out following Operation：

Pixel-matrix information using original image code tree unit calculates the edge strength of every bit, uses ES_kTo represent The edge strength of (i, j) position in N × N encoding block, its computing formula is：

Wherein, k=i × N+j, eh_kFor its horizontal component, ev_kFor its vertical component, work as m, during n ∈ { 0,1 ..., N-1 }, C_m,nRepresent the pixel value at position (m, n) place in current N × N encoding block, as m ∈ { -1, N } or n ∈ { -1, N }, C_m,nRepresent it Neighborhood territory pixel.

In calculating, because each of code tree unit of original image pixel, one 4 × 4 can be belonged simultaneously to, one 8 × 8, one 16 × 16 and 32 × 32 encoding blocks, and no matter it belongs to coded block size, its edge strength is all certain , so should be noted that the edge strength of same pixel only calculates one time, remaining is realized with position mapping mode, thus avoiding Compute repeatedly, save resources.

Step 102：The encoding block of all N × N is carried out with the classification based on the analysis of encoding block edge strength, N ∈ 4,8, 16,32}.

In this step, the encoding block of all N × N is classified successively according to the following step：

The first step, for the encoding block of each N × N, the edge strength of its all pixels point is projected according to its edge direction 33 PU model prediction directions to HEVC, find out the maximum direction of each N × N encoding block edge strength projection, as this volume The principal direction of code block；

Second step, according to the classification of encoding block principal direction, totally 4 class：

According to encoding block principal direction and the level interval D pre-setting₀:[7,14), vertically interval D₁:[23,30), minus 45 The interval D of degree₂:[14,23), other interval D₃, judge which interval encoding block falls in；

3rd step, the homogeney according to the projection of encoding block edge strength is classified, totally 2 class：

Calculation code block principal direction and its neighbouring four direction edge strength projection sum, are designated as σ；Calculate this encoding block The edge strength projection sum in all 33 directions, is designated as Σ；Judge whether following formula is set up：

σ/Σ ＞ 1-0.1 × log₂N,

If above formula is true, the edge strength projection of this encoding block is referred to as homogeneity, otherwise referred to as heterogeneous；

4th step, is classified according to pixel edge strength maximum in encoding block, totally 7 class：

Find out the maximum of pixel edge strength in encoding blockAccording to class set in advance The class interval { 0,400,900,1600,2500,4900,8100, ∞ } of separation composition, judges which kind of encoding block belongs to, Wherein, the adjacent interval folded by class separation of each two is class interval, totally seven class intervals；

5th step, second step, the 3rd step and the 4th step constitute altogether 56 classes, to each encoding block, successively according to step Rapid one, step 2, step 3 and step 4 find corresponding classification sequence number in 56 classes.

Step 103：Corresponding residual error is selected to estimate that model carries out pixel to each N × N encoding block according to classification results pre- Survey estimation error.

In this step, if the classification sequence of certain N × N encoding block is l, l ∈ { 0,1 ..., 55 }, pixel predicated error Estimate isQuantization step is Qs, and pixel edge strength is ES_k, then the corresponding residual error of this block estimate model be：

Wherein,For the parameter of N × N encoding block l class model, model ginseng The study of number, occurs before encoder design, with standard code algorithm as reference, for each point edge under each (l, N) classification Intensity level is sampled with predicated error, carries out linear fit with least square method, finally obtains under each classification mode (l, N) Parameter value；

Wherein, quantization step Qs is only relevant with coder parameters Qp, is shown below：

Wherein,Represent downward floor operation, % represents modulo operation.

Step 104：Pixel predicated error according to each N × N encoding block is estimated, the entirety calculating each N × N encoding block is compiled Code cost RD_N.

In this step, estimate the Coding cost of each N × N block according to pixel predicated error respectively, be designated as RD_N, N herein ∈ { 4,8,16,32 }, computing formula is as follows：

Wherein, w_rFor bit rate weight parameter, its value withTri- amounts of Qs, N are relevant, when encoder is realized to search The mode of table is realized, and look-up table is as follows：

And w_dFor distortion rate weight parameter, it is shown below：

Step 105：When N ∈ { 8,32 }, calculate each N × N code block segmentation Coding cost, that is, be divided into 4 correspondingCost required for encoding block, is designated as

In this step, calculate each N × N code block segmentation corresponding for 4Cost required for encoding block, note ForN ∈ { 8,32 } herein,

Wherein,Represent 4 that N × N code block segmentation is becomeEncoding block binary encoding Cost is sued for peace；γ_mode=4, represent the introduced cost of bit that partition encoding increases, γ_cbf=1, it is to represent coded block mark Introduced cost.

Step 106：When N ∈ { 8,32 }, compare the RD of each N × N encoding block_NWithValue, ifHold Row step 107；Otherwise execution step 108.

Step 107：This N × N encoding block is not split.

Step 108：This N × N code block segmentation is corresponding for 4Encoding block.

In this step 106-108, compare the RD of each N × N encoding block during N ∈ { 8,32 }_NWithValue, ifThen this encoding block is not split, otherwise, this code block segmentation is corresponding for 4Encoding block；For instructing Next code, this judgement is recorded as result, each 32 × 32 pieces of information whether split charge to big encoding block (LCB, Large Code Block) pre-segmentation Mark Array, each 8 × 8 pieces of information whether split charge to lower Item block (SCB, Small Code Block) pre-segmentation Mark Array, the present invention by from 32,/16 two kinds of encoding blocks choose one kind be used as LCB（Large Code Block）, it is used as SCB (Small Code Block) by choosing one kind from 8/4 two kinds of encoding blocks, thus permissible Make original 4 parallel encoding be changed into 2 parallel encodings, therefore decrease area and the hardware generations such as power consumption of parallel encoder Valency, shows as reducing the huge RDO modular concurrent number of hardware spending：It is kept to two from four.Additionally, the method be easy to soon The various methods that fast RD estimates combine, and reduce area and the power consumption of encoder further.

It can be seen that, in embodiments of the present invention, first the code tree unit CTU of original image is divided into several N × N's Encoding block CU, N ∈ { 4,8,16,32 }, the edge strength of every bit in calculation code block CU, then to each encoding block CU according to side Edge intensity is classified, and each encoding block is selected suitable residual error estimate that model to calculate each 32 × 32,8 according to classification results The binary encoding cost of × 8 encoding block CU and code block segmentation cost, thus judging each 32 × 32 further, 8 × 8 encoding blocks are No should be divided, if 32 × 32,8 × 8 encoding blocks all should be split, then coding work below only need to pay close attention to 16 × 16 and 4 × 4 volume Code block, if one of only 32 × 32 or 8 × 8 encoding blocks should be split, only need to pay close attention to 16 × 16 and 8 × 8 encoding blocks, or 32 × 32 and 4 × 4 encoding blocks are so that the number of encoding block CU reduces half, thus the mode combinations number of CU/PU/TU subtracts Less so that encoder complexity reduces, choose a kind of in the encoding block additionally, due to last selection 32 × 32,16 × 16, and 8 × 8, One of 4 × 4 encoding block, thus be suitable to two groups of pattern search engine Parallel Implementation, thus stably accelerating coding Process.Further, since the number of encoding block CU, before CTU coding, is first reduced by method proposed by the present invention, thus without making Become coding delay, such that it is able to realize real-time coding.

Embodiment 2：

Embodiments of the invention 2 also proposed a kind of intraframe coding based on pattern pretreatment and optimize system, referring to Fig. 2, This system includes：

Edge strength computing module 201, for being divided into the coding of several N × N by the code tree unit CTU of original image Block CU, N ∈ { 4,8,16,32 }, the edge strength of every bit in calculation code block；

Encoding block sort module 202, for carrying out to the encoding block of all N × N based on the analysis of encoding block edge strength Classification, N ∈ { 4,8,16,32 }；

Block prediction estimation error module 203, for selecting corresponding residual error to estimate model to each N according to classification results × N encoding block carries out pixel predicated error estimation, N ∈ { 4,8,16,32 }；

Binary encoding cost computing module 204, estimates for the pixel predicated error according to each N × N encoding block, calculates Binary encoding cost RD of each N × N encoding block_N, N ∈ { 4,8,16,32 }；

Partition encoding cost computing module 205, for calculating each N × N code block segmentation Coding cost, that is, is divided into 4 CorrespondingCost required for encoding block, is designated as

Whether split determination module 206, for when N ∈ { 8,32 }, the RD of relatively each N × N encoding block_NWithValue, IfThen this N × N encoding block is not split, otherwise, this N × N code block segmentation is corresponding for 4Compile Code block.

Described edge strength computing module 201 is used for the code tree unit of original image is divided into the coding of several N × N Block, N ∈ { 4,8,16,32 }, the edge strength calculating (i, j) position in the encoding block of N × N is ES_k,

Wherein, k=i × N+j, eh_kFor its horizontal component, ev_kFor its vertical component, work as m, during n ∈ { 0,1 ..., N-1 }, C_m,nRepresent the pixel value at position (m, n) place in current N × N encoding block, as m ∈ { -1, N } or n ∈ { -1, N }, C_m,nRepresent it Neighborhood territory pixel value.

Described encoding block sort module 202 includes：

Projection subelement 2020, for the encoding block for each N × N, by the edge strength of its all pixels point according to it 33 PU model prediction directions in edge direction projection paramount efficiency video coding HEVC, find out each N × N coding block edge strong The maximum direction of degree projection, as the principal direction of this encoding block；

First classification subelement 2021, for according to the classification of encoding block principal direction, totally 4 class：

According to encoding block principal direction and the level interval D pre-setting₀:[7,14), vertically interval D₁:[23,30), minus 45 The interval D of degree₂:[14,23), other interval D₃, judge which interval encoding block falls in；

Second classification subelement 2022, for being classified according to the homogeney of encoding block edge strength projection, totally 2 class：

Calculation code block principal direction and its neighbouring four direction edge strength projection sum, are designated as σ；Calculate this encoding block The edge strength projection sum in all 33 directions, is designated as Σ；Judge whether following formula is set up：

σ/Σ ＞ 1-0.1 × log₂N,

If above formula is true, the edge strength projection of this encoding block is referred to as homogeneity, otherwise referred to as heterogeneous；

3rd classification subelement 2023, for being classified according to pixel edge strength maximum in encoding block, totally 7 Class：

Find out the maximum of pixel edge strength in encoding blockAccording to set in advance The class interval { 0,400,900,1600,2500,4900,8100, ∞ } of class separation composition, judges which encoding block belongs to Class, wherein, the adjacent interval folded by class separation of each two is class interval, totally seven class intervals；

Classification subelement 2024, for each encoding block, successively according to projection subelement, the first classification subelement, Second classification subelement and the 3rd classification subelement are from the first classification subelement, the second classification subelement and the 3rd classification subelement Corresponding classification sequence number is found in 56 kinds of classification results of composition.

If the classification sequence that described block prediction estimation error module 203 is used for a certain N × N encoding block is l, quantify step A length of Qs, pixel edge strength is ES_k, calculate pixel predicated error estimate

Wherein, a (l, N), b_k(l,N),k∈{0,1,...,N²- 1 } for the ginseng of l class N × N encoding block residual prediction model Number, l ∈ { 0,1 ..., 55 }.

Described binary encoding cost computing module 204 is used for, when N ∈ { 4,8,16,32 }, calculating each N × N encoding block Binary encoding cost RD_N：

Wherein, w_rFor bit rate weight parameter, its value withTri- amounts of Qs, N are relevant, when encoder is realized to search The mode of table is realized, w_dFor distortion rate weight parameter, it is shown below：

Described partition encoding cost computing module 205 is used for, when N ∈ { 8,32 }, calculating each N × N code block segmentation coding Cost, that is, be divided into 4 correspondingCost required for encoding block is designated as

Wherein,Represent 4 that N × N code block segmentation is becomeEncoding block binary encoding Cost is sued for peace；γ_mode=4, represent the introduced cost of bit that partition encoding increases, γ_cbf=1, it is to represent coded block mark Introduced cost.

In embodiments of the present invention, calculate edge strength first with edge strength computing module, then utilize encoding block Sort module is classified according to edge strength to each encoding block, recycles block prediction estimation error module, binary encoding Cost computing module and partition encoding cost computing module calculate respectively each 32 × 32,8 × 8 encoding blocks binary encoding cost and Code block segmentation Coding cost, finally judges whether it should be divided further using whether splitting determination module；If 32 × 32, 8 × 8 encoding blocks all should be split, then coding work below only need to pay close attention to 16 × 16 and 4 × 4 encoding blocks, if only 32 × 32 or 8 One of × 8 encoding blocks should be split, then only need to pay close attention to 16 × 16 and 8 × 8 encoding blocks, or 32 × 32 and 4 × 4 encoding blocks, The number so making encoding block CU reduces half, thus the mode combinations number of CU/PU/TU reduces so that encoder complexity drops Low, choose a kind of in the encoding block additionally, due to last selection 32 × 32,16 × 16, and 8 × 8, in 4 × 4 encoding block Kind, thus being suitable to two groups of pattern search engine Parallel Implementation, thus stably accelerating cataloged procedure.Further, since this The system of bright proposition was applied before CTU encodes, and first reduced the number of encoding block CU, thus without causing coding delay, from And real-time coding can be realized.

In the embodiment of the present invention, a kind of due to choosing in the encoding block of last selection 32 × 32,16 × 16, and 8 × 8,4 × One of 4 encoding block, thus being suitable to two groups of pattern search engine Parallel Implementation, showing as in actual coding application can To reduce area and the hardware costs such as power consumption of parallel encoder, that is, reduce the huge RDO modular concurrent number of hardware spending：From Four are kept to two.Additionally, the various methods that the method is easy to estimate with quick RD combine, reduce the face of encoder further Amass and power consumption.

Above example is merely to illustrate technical scheme, is not intended to limit；Although with reference to the foregoing embodiments The present invention has been described in detail, it will be understood by those within the art that：It still can be to aforementioned each enforcement Technical scheme described in example is modified, or carries out equivalent to wherein some technical characteristics；And these are changed or replace Change, do not make the essence of appropriate technical solution depart from the spirit and scope of various embodiments of the present invention technical scheme.

Claims (10)

1. a kind of intraframe coding optimization method based on pattern pretreatment is it is characterised in that the method includes：

S1, the code tree unit of original image is divided into the encoding block of several N × N, N ∈ { 4,8,16,32 }, calculation code block The edge strength of interior each pixel；

S2, the encoding block to all N × N carry out the classification based on the edge strength analysis of all pixels point in encoding block, N ∈ {4,8,16,32}；

S3, corresponding residual error is selected to estimate that model carries out pixel predicated error to each N × N encoding block and estimates according to classification results Meter, N ∈ { 4,8,16,32 }；

S4, the pixel predicated error estimation according to each N × N encoding block, calculate binary encoding cost RD of each N × N encoding block_N, N∈{4,8,16,32}；

S5, when N ∈ { 8,32 }, calculate each N × N code block segmentation Coding cost, that is, be divided into 4 correspondingCoding Cost required for block, is designated as

${\mathrm{RD}}_{\⊗N}=\underset{n=0}{\overset{3}{\Σ}}{\mathrm{RD}}_{N/2}\left(n\right)+3\×\frac{7}{64}\×({\mathrm{\γ}}_{\mathrm{mod}e}+{\mathrm{\γ}}_{cbf});$

Wherein,Represent 4 that N × N code block segmentation is becomeEncoding block binary encoding cost Summation；γ_mode=4, represent the introduced cost of bit that partition encoding increases, γ_cbf=1, it is to represent that coded block mark is drawn Enter cost；

S6, when N ∈ { 8,32 }, the RD of relatively each N × N encoding block_NWithValue, ifThen this N × N coding Block is not split, otherwise, this N × N code block segmentation is corresponding for 4Encoding block.

2. method according to claim 1 is it is characterised in that described step S1 includes：The code tree of original image is single Unit is divided into the encoding block of several N × N, N ∈ { 4,8,16,32 }, calculates the edge strength of (i, j) position in the encoding block of N × N For ES_k,

$\left\{\begin{array}{c}{\mathrm{ES}}_k={\mathrm{eh}}_k^2+{\mathrm{ev}}_k^2\\ {\mathrm{eh}}_k=C_{i-1,j}-2\×C_{i,j}+C_{i+1,j}\\ {\mathrm{ev}}_k=C_{i,j-1}-2\×C_{i,j}+C_{i,j+1}\end{array}\right.,$

Wherein, k=i × N+j, eh_kFor its horizontal component, ev_kFor its vertical component, work as m, during n ∈ { 0,1 ..., N-1 }, C_m,n Represent the pixel value at position (m, n) place in current N × N encoding block, as m ∈ { -1, N } or n ∈ { -1, N }, C_m,nRepresent that it is adjacent Domain pixel value.

3. method according to claim 1 is it is characterised in that described step S2 includes：

S21, the encoding block for each N × N, the edge strength of its all pixels point is projected to high efficiency according to its edge direction 33 PU model prediction directions in Video coding HEVC, find out the maximum direction of each N × N encoding block edge strength projection, make Principal direction for this encoding block；

S22, foundation encoding block principal direction classification, totally 4 class：

According to encoding block principal direction and the level interval D pre-setting₀:[7,14), vertically interval D₁:[23,30), Fu45Du area Between D₂:[14,23), other interval D₃, judge which interval encoding block falls in；

S23, the homogeney of foundation encoding block edge strength projection are classified, totally 2 class：

Calculation code block principal direction and its neighbouring four direction edge strength projection sum, are designated as σ；Calculate this encoding block to own The edge strength projection sum in 33 directions, is designated as Σ；Judge whether following formula is set up：

σ/Σ>1-0.1×log₂N,

If above formula is true, the edge strength projection of this encoding block is referred to as homogeneity, otherwise referred to as heterogeneous；

S24, classified according to pixel edge strength maximum in encoding block, totally 7 class：

Find out the maximum of pixel edge strength in encoding blockAccording to class set in advance boundary The class interval { 0,400,900,1600,2500,4900,8100, ∞ } of point composition, judges which kind of encoding block belongs to, wherein, The adjacent interval folded by class separation of each two is class interval, totally seven class intervals；

S25, step S22, S23 and S24 constitute altogether 56 classes, to each encoding block, successively according to step S21, S22, S23, S24 find corresponding classification sequence number in 56 classes.

4. method according to claim 1 is it is characterised in that described step S3 includes：

If the classification sequence of a certain N × N encoding block is l, quantization step is Qs, and pixel edge strength is ES_k, calculate this coding The pixel predicated error estimate of block

Wherein, a (l, N), b_k(l,N),k∈{0,1,...,N²- 1 } for the parameter of l class N × N encoding block residual prediction model, l ∈{0,1,...,55}.

5. method according to claim 1 is it is characterised in that described step S4 includes：

When N ∈ { 4,8,16,32 }, binary encoding cost RD of each N × N encoding block_NComputing formula is as follows：

Wherein,For pixel predicated error estimate, Qs is quantization step, w_rFor bit rate weight parameter, its value with Tri- amounts of Qs, N are relevant, realized in the way of look-up table when encoder is realized, w_dFor distortion rate weight parameter, as following formula institute Show：

6. a kind of intraframe coding based on pattern pretreatment optimizes system it is characterised in that this system includes：

Edge strength computing module, for the code tree unit of original image being divided into the encoding block of several N × N, N ∈ 4, 8,16,32 }, in calculation code block each pixel edge strength；

Encoding block sort module, for carrying out the edge strength based on all pixels point in encoding block to the encoding block of all N × N The classification of analysis, N ∈ { 4,8,16,32 }；

Block prediction estimation error module, for selecting corresponding residual error to estimate that model encodes to each N × N according to classification results Block carries out pixel predicated error estimation, N ∈ { 4,8,16,32 }；

Binary encoding cost computing module, estimates for the pixel predicated error according to each N × N encoding block, calculates each N × N Binary encoding cost RD of encoding block_N, N ∈ { 4,8,16,32 }；

Partition encoding cost computing module, for calculating each N × N code block segmentation Coding cost, that is, be divided into 4 correspondingCost required for encoding block, is designated as

${\mathrm{RD}}_{\⊗N}=\underset{n=0}{\overset{3}{\Σ}}{\mathrm{RD}}_{N/2}\left(n\right)+3\×\frac{7}{64}\×({\mathrm{\γ}}_{\mathrm{mod}e}+{\mathrm{\γ}}_{cbf});$

Wherein,Represent 4 that N × N code block segmentation is becomeEncoding block binary encoding cost Summation；γ_mode=4, represent the introduced cost of bit that partition encoding increases, γ_cbf=1, it is to represent that coded block mark is drawn Enter cost；

Whether split determination module, for when N ∈ { 8,32 }, the RD of relatively each N × N encoding block_NWithValue, ifThen this N × N encoding block is not split, otherwise, this N × N code block segmentation is corresponding for 4Coding Block.

7. system according to claim 6 is it is characterised in that described edge strength computing module is used for original image Code tree unit is divided into the encoding block of several N × N, N ∈ { 4,8,16,32 }, calculates (i, j) position in the encoding block of N × N Edge strength is ES_k：

$\left\{\begin{array}{c}{\mathrm{ES}}_k={\mathrm{eh}}_k^2+{\mathrm{ev}}_k^2\\ {\mathrm{eh}}_k=C_{i-1,j}-2\×C_{i,j}+C_{i+1,j}\\ {\mathrm{ev}}_k=C_{i,j-1}-2\×C_{i,j}+C_{i,j+1}\end{array}\right.,$

Wherein, k=i × N+j, eh_kFor its horizontal component, ev_kFor its vertical component, when

During m, n ∈ { 0,1 ..., N-1 }, C_m,nRepresent the pixel value at position (m, n) place in current N × N encoding block, when m ∈ -1, N } or during n ∈ { -1, N }, C_m,nRepresent its neighborhood territory pixel value.

8. system according to claim 6 is it is characterised in that described encoding block sort module includes：

Projection subelement, for the encoding block for each N × N, by the edge strength of its all pixels point according to its edge direction Project 33 PU model prediction directions in paramount efficiency video coding HEVC, find out each N × N encoding block edge strength and project Big direction, as the principal direction of this encoding block；

First classification subelement, for according to the classification of encoding block principal direction, totally 4 class：

According to encoding block principal direction and the level interval D pre-setting₀:[7,14), vertically interval D₁:[23,30), Fu45Du area Between D₂:[14,23), other interval D₃, judge which interval encoding block falls in；

Second classification subelement, for being classified according to the homogeney of encoding block edge strength projection, totally 2 class：

Calculation code block principal direction and its neighbouring four direction edge strength projection sum, are designated as σ；Calculate this encoding block to own The edge strength projection sum in 33 directions, is designated as Σ；Judge whether following formula is set up：

σ/Σ>1-0.1×log₂N,

If above formula is true, the edge strength projection of this encoding block is referred to as homogeneity, otherwise referred to as heterogeneous；

3rd classification subelement, for being classified according to pixel edge strength maximum in encoding block, totally 7 class：

Find out the maximum of pixel edge strength in encoding blockAccording to class set in advance boundary The class interval { 0,400,900,1600,2500,4900,8100, ∞ } of point composition, judges which kind of encoding block belongs to, wherein, The adjacent interval folded by class separation of each two is class interval, totally seven class intervals；

Classification subelement, for each encoding block, successively according to projection subelement, the first classification subelement, the second classification Subelement and the 3rd classification subelement from first classify subelement, second classification subelement and the 3rd classification subelement composition 56 Plant in classification results and find corresponding classification sequence number.

9. system according to claim 6 is it is characterised in that described block prediction estimation error module is used for when a certain The classification sequence of N × N encoding block is l, and quantization step is Qs, and pixel edge strength is ES_kWhen, calculate pixel predicated error Estimate

Wherein, a (l, N), b_k(l,N),k∈{0,1,...,N²- 1 } for the parameter of l class N × N encoding block residual prediction model, l ∈{0,1,...,55}.

10. system according to claim 6 is it is characterised in that described binary encoding cost computing module is used for as N ∈ When { 4,8,16,32 }, calculate binary encoding cost RD of each N × N encoding block_N：

Wherein,For pixel predicated error estimate, Qs is quantization step, w_rFor bit rate weight parameter, its value with Tri- amounts of Qs, N are relevant, realized in the way of look-up table when encoder is realized, w_dFor distortion rate weight parameter, as following formula institute Show：