CN106131554A - The HEVC point self-adapted compensation method of quick sample product based on major side direction - Google Patents

Abstract

The invention discloses a kind of HEVC point self-adapted compensation method of quick sample product based on major side direction.The present invention comprises the steps: step 1, extracts the major side direction of each CTU；Step 2, the Optimal Boundary compensation model determining CTU and offset thereof；Step 3, calculate the relative rate distortion cost of other SAO compensation ways；Step 4, determine the optimum SAO pattern of CTU.The present invention extracts the major side direction of each CTU and as the optimum EO pattern of this CTU, then travels through BO pattern, Parameter fusion pattern, uncompensation pattern, rules out optimization model by formula Cost=Distortion+ λ × Bitrate.The present invention utilizes the texture information of image to reduce the complexity of SAO mode determination process, and the SAO saving HEVC encodes the operation time.

Description

The HEVC point self-adapted compensation method of quick sample product based on major side direction

Technical field

The invention belongs to high-definition digital video compression coding and decoding technical field, particularly relate to a kind of based on major side direction HEVC (High Efficiency Video Coding, efficient video coding) the point self-adapted compensation method of quick sample product.

Background technology

In the fast development of digital video application industrial chain, in the face of Video Applications is constantly to fine definition, high frame per second, height The trend that compression ratio direction is developed, previous generation video compression standard agreement limitation H.264/AVC constantly highlights.Therefore, by International Telecommunication Union (ITU) and Motion Picture Experts Group (MPEG) combine the tissue JCTVC of establishment and propose video volume of future generation Decoding standard is H.265/HEVC.Its target is under identical picture quality premise, and compression ratio is than H.264/AVC high-grade Double.

HEVC does not break through in terms of coding principle and basic structure, and the most basically identical, i.e. prediction adds conversion Block encoding mode；On Ciphering details and the most also very close to, comprise infra-frame prediction, inter prediction, estimation With coding/decoding modules such as compensation, orthogonal transformation, quantization, loop filtering, entropy code and reconstructions.But, and H.264/AVC compare Relatively, HEVC almost takes important corrective measure in each coding link, as infra-frame prediction supports 35 kinds of patterns, frames Between prediction introduce that Merge pattern, change quantization support are up to 32 × 32 converter units, entropy code uses CABAC (based on context Adaptive binary arithmetic coding, Context Adaptive Binary Arithmetic Coder) algorithm etc., in addition HEVC loop filtering technology not only continues to use the de-blocking filter in H.264 standard, and adds sampling point adaptive equalization (Sample Adaptive Offset, SAO) filtering, reduces the video brought by the produced ringing effect of estimation and change quantization subjective Mass loss.

H.265/HEVC the SAO in standard is with CTU as ultimate unit, by selecting a suitable grader will rebuild picture Element divides classification, then different classes of pixel is used different offsets, can be effectively improved the subjective and objective quality of video.It Including two large compensation forms, it is boundary compensation (Edge Offset, EO) and sideband compensation (Band Offset, BO) respectively, this Also introduce outward Parameter fusion technology.

Boundary compensation pattern uses a kind of 3 dot structures to classify handled pixel.On limit residing for this pixel Add 2 pixels of arest neighbors on the direction of edge (one-dimensional), centered by this pixel, form level (EO_0), vertical (EO_ 1), 135 ° (EO_2), 45 ° (EO_3) four kind of 3 dot structure, as shown in Figure 1.According to the different distributions of 3 pixel values, this pixel can Being divided into peak pixel (if its value is more than 2 adjacent pixels), valley pixel is (if its value is less than 2 adjacent pictures Element), edge pixel (if its value is equal to any one neighbor) and other pixel (if not meeting first 3 kinds), totally 4 class, As shown in Figure 2.Deviant pixel added by different to 4 classes is the most different.

Sideband compensation technique is sorted out according to pixel intensity value, and pixel coverage is divided into 32 sidebands, Mei Gebian by it Band compensates according to self pixel feature, and same sideband uses identical offset.HEVC standard one CTB of regulation can only Select 4 continuous print sidebands, and only the pixel belonging to these 4 sidebands is compensated.

Parameter fusion (Merge) refers to that, for a CTB block, its SAO parameter directly uses the SAO parameter of adjacent block, this Time have only to identify the SAO parameter that have employed which adjacent block.

To determine the SAO parameter of a CTU block, need to travel through 4 kinds of EO patterns, BO pattern, Merge pattern and uncompensation Pattern, computationally intensive.

Summary of the invention

The present invention is directed to technical problem present in above-mentioned prior art, utilize the texture information of image, for each CTB, it is proposed that a kind of HEVC point self-adapted compensation method of quick sample product based on major side direction, the method includes walking as follows Rapid:

Step 1, extract the major side direction of each CTU (tree-like coding unit, coding tree unit).

In order to utilize the texture information of image, the absolute value sum of each CTU pixel difference in side the most in the same direction to be calculated, Defining this absolute value sum is major side orientation consistency (Dominant Edge Assent, DEA), candidate major side side To including 0 °, 45 °, 90 °, 135 ° of four directions, and the concordance of the four direction of correspondence is expressed as DEA₁、DEA₂、 DEA₃、DEA₄, and take the major side direction as this CTU, the direction corresponding to concordance minima；Extract the master of each CTU Edge direction is wanted to specifically comprise the following steps that

The CTU block of N × N (N is generally 64) size is divided into 5 sub-blocks by 1-1., and calculates the pixel of 5 sub-blocks respectively Meansigma methods.The CTU block of N × N size is divided into 5 sub-blocks of a, b, c, d, e, calculates the average of 5 all pixels of sub-block respectively Value P_a、P_b、P_c、P_d、P_eAs follows:

$P_a=\underset{i=0}{\overset{i=\frac{N}{2}-1}{\Σ}}\underset{j=0}{\overset{j=\frac{N}{2}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_b=\underset{i=\frac{N}{2}}{\overset{i=N-1}{\Σ}}\underset{j=0}{\overset{j=\frac{N}{2}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_c=\underset{i=0}{\overset{i=\frac{N}{2}-1}{\Σ}}\underset{j=\frac{N}{2}}{\overset{j=N-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_d=\underset{i=\frac{N}{2}}{\overset{i=N-1}{\Σ}}\underset{j=\frac{N}{2}}{\overset{j=N-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_e=\underset{i=\frac{N}{4}}{\overset{i=\frac{3N}{4}-1}{\Σ}}\underset{j=\frac{N}{4}}{\overset{j=\frac{3N}{4}-1}{\Σ}}{P}_{ij}\frac{N\×N}{4}$

Wherein, P_ijFor CTU at (i, j) pixel value of position.

1-2. utilizes P_a、P_b、P_c、P_d、P_eCalculate 4 candidate major side orientation consistency DEA respectively₁、DEA₂、DEA₃、 DEA₄:

DEA₁=| P_b-P_a|+|P_d-P_c|

DEA₂=| P_c-P_e|+|P_e-P_b|

DEA₃=| P_c-P_a|+|P_d-P_b|

DEA₄=| P_d-P_e|+|P_e-P_a|

1-3. is by DEA₁、DEA₂、DEA₃、DEA₄Candidate's major side direction corresponding to middle minima is as the master of this CTU Wanting edge direction DE, DE i.e. represents 0 °, 45 °, 90 °, a direction in 135 °.Conforming minima is defined as follows:

DEA_min=min{DEA₁,DEA₂,DEA₃,DEA₄}

When to Image semantic classification, if image resolution ratio is not the multiple (N is generally 64) of N, can be with 0 pixel to image Edge is filled with, and in the present invention, the CTU filled is referred to as imperfect CTU, and for imperfect CTU, it is calculated as above-mentioned The texture feature of image can not be reflected in the major side direction gone out, and according to image space dependency, makes its major side direction be The major side direction of nearest encoded CTU.

Step 2, Optimal Boundary compensation (EO) pattern determining CTU and offset thereof.

The optimum EO pattern of CTU is determined by step 1 calculated major side direction, i.e. DEA₁、DEA₂、DEA₃、DEA₄ Corresponding EO pattern is respectively as follows: EO_0, EO_1, EO_2, EO_3, then calculates its offset according to mode type.

After determining pattern, the pixel in CTU is divided into 5 classes, and for belonging to 1～4 classes, its pixel needs to mend Repaying, each kind can select again different offsets, it is therefore desirable to chooses optimal compensation value M for each kind.Each kind The determination process of optimal compensation value is independently carried out.

It is original pixels and reconstructed pixel (before SAO compensates) that 2-1. utilizes statistical information to calculate initial compensation value m=E/N, E Between difference sum, N is the pixel number belonging to this kind, and m rounds the offset m ' limit after obtaining m ' and rounding System is between [-7,7]；

2-2. limits adjustment according to boundary compensation kind further to offset, for kind 1 and kind 2, compensates Value have to be larger than equal to 0, and the offset of kind 3 and kind 4 is necessarily less than equal to 0.

2-3. travels through all of candidate's offset [0, m '] or [m ', 0], chooses the offset that relative rate distortion cost is minimum As optimal compensation value M, now number of coded bits R=| M |+1.Rate distortion costs Cost is defined as

Cost=Distortion+ λ × Bitrate

Original image and the mistake reconstructing image (after SAO compensates) under wherein Distortion (distortion) is a certain EO pattern Very, λ (lambda) is Lagrange factor, and Bitrate (bit rate) is the bit number needed for coding SAO parameter information.

Step 3, calculate the relative rate distortion cost of other SAO compensation ways.

In order to determine the optimum SAO pattern of CTU, also need to calculate sideband compensation (BO) pattern relative with Parameter fusion pattern Rate distortion costs.

The determination process of BO pattern:

First, it is followed successively by 0～31 sideband and chooses optimal compensation value；

Then, it is determined that continuous 4 sidebands of optimum, the determination process of single sideband offset is as follows: utilize statistical information Calculating initial compensation value m1=E/N1, E is the difference sum between original pixels and reconstructed pixel (before SAO compensates), and N1 is for belonging to In the pixel number of this sideband, m1 is rounded and obtains m1 ', and the offset m1 ' after rounding is limited between [-7,7]；? M1 ' value after amplitude limit；

Subsequently, determine that candidate's offset scope is for [0, m1 '] or [m1 ', 0] according to its symbol.

Finally, travel through all candidate's offsets, choose the minimum offset of relative rate distortion cost as optimal compensation value M1, now number of coded bits R=| M1 |+2.Choose 4 continuous sidebands so that rate distortion costs is minimum and be optimum BO mould Formula, the offset sum that 4 sidebands are corresponding is optimal compensation value.

In the rate distortion computation formula of Parameter fusion pattern, Distortion by employing parameter adjacent block relative rate lose True cost, Bitrate is flag bit number of coded bits.

Step 4, determine the optimum SAO pattern of CTU.

The compensation type of SAO includes uncompensation, EO pattern, BO pattern, left adjacent block Parameter fusion pattern and upper adjacent block Parameter fusion pattern, compensation type final for SAO selects from 5 kinds of patterns, one group of SAO ginseng that selection rate distortion performance is optimum Number.

So far, the whole HEVC point self-adapted compensation method of quick sample product based on major side direction terminates.

The present invention has the beneficial effect that:

This method essence utilizes the texture information of image to reduce the complexity of SAO mode determination process, thus reduces HEVC Scramble time.Specifically, this method major side direction by each CTU of extraction the optimum EO as this CTU Pattern, then traversal BO pattern, Parameter fusion pattern, uncompensation pattern, by formula Cost=Distortion+ λ × Bitrate Rule out optimization model, compared to traditional method traveling through all EO patterns, can be big in the situation that encoding efficiency loss is little Amount saves the SAO scramble time of HEVC.

Accompanying drawing explanation

4 kinds of boundary compensation pattern diagram of Fig. 1: SAO.

The classification schematic diagram of Fig. 2: boundary compensation.

Fig. 3: CTU major side direction schematic diagram.

Fig. 4: CTU divides schematic diagram.

Fig. 5: HEVC quick sample product based on major side direction point self-adapted compensation method flow chart.

Detailed description of the invention

As a example by the cataloged procedure of several sequences such as Basketball Drill, for reducing ringing effect, it need to be carried out Sampling point adaptive equalization, the inventive method proposed is described in detail by the present embodiment accordingly.

The step of the HEVC point self-adapted compensation method of quick sample product based on major side direction is as follows:

Step 1, extract the major side direction of each CTU (tree-like coding unit, coding tree unit).

In order to utilize the texture information of image, the absolute value sum of each CTU pixel difference in side the most in the same direction to be calculated, Defining this absolute value sum is major side orientation consistency (Dominant Edge Assent, DEA), candidate major side side To including 0 °, 45 °, 90 °, 135 ° of four directions, as it is shown on figure 3, and the concordance of four direction of correspondence be expressed as DEA₁、DEA₂、DEA₃、DEA₄, and take the major side direction as this CTU, the direction corresponding to concordance minima；Extract every The major side direction of individual CTU specifically comprises the following steps that

The CTU block of N × N (N is generally 64) size is divided into 5 sub-blocks by 1-1., and calculates the pixel of 5 sub-blocks respectively Meansigma methods.As shown in Figure 4 the CTU block of N × N size is divided into 5 sub-blocks of a, b, c, d, e, calculates 5 sub-blocks respectively and own The average value P of pixel_a、P_b、P_c、P_d、P_eAs follows:

$P_a=\underset{i=0}{\overset{i=\frac{N}{2}-1}{\Σ}}\underset{j=0}{\overset{j=\frac{N}{2}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_b=\underset{i=\frac{N}{2}}{\overset{i=N-1}{\Σ}}\underset{j=0}{\overset{j=\frac{N}{2}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_c=\underset{i=0}{\overset{i=\frac{N}{2}-1}{\Σ}}\underset{j=\frac{N}{2}}{\overset{j=N-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_d=\underset{i=\frac{N}{2}}{\overset{i=N-1}{\Σ}}\underset{j=\frac{N}{2}}{\overset{j=N-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_e=\underset{i=\frac{N}{4}}{\overset{i=\frac{3N}{4}-1}{\Σ}}\underset{j=\frac{N}{4}}{\overset{j=\frac{3N}{4}-1}{\Σ}}{P}_{ij}\frac{N\×N}{4}$

Wherein, P_ijFor CTU at (i, j) pixel value of position.

1-2. utilizes P_a、P_b、P_c、P_d、P_eCalculate 4 candidate major side orientation consistency DEA respectively₁、DEA₂、DEA₃、 DEA₄:

DEA₁=| P_b-P_a|+|P_d-P_c|

DEA₂=| P_c-P_e|+|P_e-P_b|

DEA₃=| P_c-P_a|+|P_d-P_b|

DEA₄=| P_d-P_e|+|P_e-P_a|

1-3. is by DEA₁、DEA₂、DEA₃、DEA₄Candidate's major side direction corresponding to middle minima is as the master of this CTU Wanting edge direction DE, DE i.e. represents 0 °, 45 °, 90 °, a direction in 135 °.Conforming minima is defined as follows:

DEA_min=min{DEA₁,DEA₂,DEA₃,DEA₄}

When to Image semantic classification, if image resolution ratio is not the multiple (N is generally 64) of N, can be with 0 pixel to image Edge is filled with, and in the present invention, the CTU filled is referred to as imperfect CTU, and for imperfect CTU, it is calculated as above-mentioned The texture feature of image can not be reflected in the major side direction gone out, and according to image space dependency, makes its major side direction be The major side direction of nearest encoded CTU.

Step 2, Optimal Boundary compensation (EO) pattern determining CTU and offset thereof.

The optimum EO pattern of CTU is determined by step 1 calculated major side direction, i.e. DEA₁、DEA₂、DEA₃、DEA₄ Corresponding EO pattern is respectively as follows: EO_0, EO_1, EO_2, EO_3, then calculates its offset according to mode type.

After determining pattern, the pixel in CTU is divided into 5 classes, i.e. 4 classes shown in Fig. 2 and is not belonging to the 5th class of this 4 class Situation, for belonging to 1～4 classes, its pixel needs to compensate, and each kind can select again different offsets, therefore Need to choose optimal compensation value M for each kind.The determination process of each kind optimal compensation value is independently carried out.

It is original pixels and reconstructed pixel (before SAO compensates) that 2-1. utilizes statistical information to calculate initial compensation value m=E/N, E Between difference sum, N is the pixel number belonging to this kind, and m rounds the offset m ' limit after obtaining m ' and rounding System is between [-7,7]；

2-2. limits adjustment according to boundary compensation kind further to offset, for kind 1 and kind 2, compensates Value have to be larger than equal to 0, and the offset of kind 3 and kind 4 is necessarily less than equal to 0.

2-3. travels through all of candidate's offset [0, m '] or [m ', 0], chooses the offset that relative rate distortion cost is minimum As optimal compensation value M, now number of coded bits R=| M |+1.Rate distortion costs Cost is defined as

Cost=Distortion+ λ × Bitrate

Original image and the mistake reconstructing image (after SAO compensates) under wherein Distortion (distortion) is a certain EO pattern Very, λ (lambda) is Lagrange factor, and Bitrate (bit rate) is the bit number needed for coding SAO parameter information.

Step 3, calculate the relative rate distortion cost of other SAO compensation ways.

In order to determine the optimum SAO pattern of CTU, also need to calculate sideband compensation (BO) pattern relative with Parameter fusion pattern Rate distortion costs.

The determination process of BO pattern:

First, it is followed successively by 0～31 sideband and chooses optimal compensation value；

Then, it is determined that continuous 4 sidebands of optimum, the determination process of single sideband offset is as follows: utilize statistical information Calculating initial compensation value m1=E/N1, E is the difference sum between original pixels and reconstructed pixel (before SAO compensates), and N1 is for belonging to In the pixel number of this sideband, m1 is rounded and obtains m1 ', and the offset m1 ' after rounding is limited between [-7,7]；? M1 ' value after amplitude limit；

Subsequently, determine that candidate's offset scope is for [0, m1 '] or [m1 ', 0] according to its symbol.

Finally, travel through all candidate's offsets, choose the minimum offset of relative rate distortion cost as optimal compensation value M1, now number of coded bits R=| M1 |+2.Choose 4 continuous sidebands so that rate distortion costs is minimum and be optimum BO mould Formula, corresponding offset is optimal compensation value.

In the rate distortion computation formula of Parameter fusion pattern, Distortion by employing parameter adjacent block relative rate lose True cost, Bitrate is flag bit number of coded bits.

Step 4, determine the optimum SAO pattern of CTU.

The compensation type of SAO includes uncompensation, EO pattern, BO pattern, left adjacent block Parameter fusion pattern and upper adjacent block Parameter fusion pattern, compensation type final for SAO selects from 5 kinds of patterns, one group of SAO ginseng that selection rate distortion performance is optimum Number.

So far, the whole HEVC point self-adapted compensation method of quick sample product based on major side direction terminates, method flow diagram As shown in Figure 5.

Experiment test environment of the present invention uses HEVC standard identifying code HM13.0 in window8 system VS2010 compiling fortune OK, 16 cycle testss are added up and tested.With full I frame (All Intra, AI), random access memory (Random Access, RA), low latency-P frame (LP), low latency-B frame (LB) be coding environment, arrange 22,27,32,37 as QP value, utilize brightness Bjontegaard-Delta bit rate (YBD-rate) sideband compensation optimizing improved method and the original SAO of HEVC that will propose Parameter determination method contrasts, and adds up the encoding and decoding time, and result is as shown in table 1.

By table 1 it appeared that the present invention proposes fast method compared with original method, save the SAO coding of 53.39% Time, and the loss of encoding efficiency is the least.

Table 1 HEVC based on the major side direction point self-adapted compensation method of quick sample product and HEVC source code Contrast on effect result

Certainly, those of ordinary skill in the art is it should be appreciated that above example is intended merely to this is described Bright, and be not intended as limitation of the invention, as long as within the scope of the invention, to the change of above example, modification all Protection scope of the present invention will be fallen into.

Claims (4)

1. the HEVC point self-adapted compensation method of quick sample product based on major side direction, it is characterised in that comprise the steps:

Step 1, extract the major side direction of each CTU；

Calculating the absolute value sum of each CTU pixel difference in side the most in the same direction, defining this absolute value sum is major side side To concordance DEA, candidate's major side direction includes 0 °, 45 °, 90 °, 135 ° of four directions, and the one of the four direction of correspondence Cause property is expressed as DEA₁、DEA₂、DEA₃、DEA₄, and take main as this CTU of the direction corresponding to concordance minima Edge direction；

Step 2, the Optimal Boundary compensation model determining CTU and offset thereof；

The optimum EO pattern of CTU is determined by step 1 calculated major side direction, i.e. DEA₁、DEA₂、DEA₃、DEA₄Corresponding EO pattern be respectively as follows: EO_0, EO_1, EO_2, EO_3, then calculate its offset according to mode type；After determining pattern, Pixel in CTU is divided into 5 classes, and for belonging to 1～4 classes, its pixel needs to compensate, and each kind can select not again Same offset, it is therefore desirable to choose optimal compensation value M for each kind；The determination process of each kind optimal compensation value is independent Carry out；

Step 3, calculate the relative rate distortion cost of other SAO compensation ways；

In order to determine the optimum SAO pattern of CTU, also need to calculate sideband and compensate the relative rate mistake of (BO) pattern and Parameter fusion pattern True cost；

Step 4, determine the optimum SAO pattern of CTU；

The compensation type of SAO includes uncompensation, EO pattern, BO pattern, left adjacent block Parameter fusion pattern and upper adjacent block parameter Fusion mode, compensation type final for SAO selects from 5 kinds of patterns, one group of SAO parameter that selection rate distortion performance is optimum.

The HEVC point self-adapted compensation method of quick sample product based on major side direction the most according to claim 1, its feature It is that the major side direction of each CTU of extraction described in step 1 specifically comprises the following steps that

The CTU block of N × N size is divided into 5 sub-blocks by 1-1., and calculates the pixel average of 5 sub-blocks respectively；N × N is big Little CTU block is divided into a, b, c, d, e5 sub-block, calculates the average value P of 5 all pixels of sub-block respectively_a、P_b、P_c、P_d、P_eAs Under:

$P_a=\underset{i=0}{\overset{i=\frac{N}{2}-1}{\Σ}}\underset{j=0}{\overset{j=\frac{N}{2}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_b=\underset{i=\frac{N}{2}}{\overset{i=N-1}{\Σ}}\underset{j=0}{\overset{j=\frac{N}{2}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_c=\underset{i=0}{\overset{i=\frac{N}{2}-1}{\Σ}}\underset{j=\frac{N}{2}}{\overset{j=N-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_d=\underset{i=\frac{N}{2}}{\overset{i=N-1}{\Σ}}\underset{j=\frac{N}{2}}{\overset{j=N-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

$P_e=\underset{i=\frac{N}{4}}{\overset{i=\frac{3N}{4}-1}{\Σ}}\underset{j=\frac{N}{4}}{\overset{j=\frac{3N}{4}-1}{\Σ}}{P}_{ij}/\frac{N\×N}{4}$

Wherein, P_ijFor CTU at (i, j) pixel value of position；

1-2. utilizes P_a、P_b、P_c、P_d、P_eCalculate 4 candidate major side orientation consistency DEA respectively₁、DEA₂、DEA₃、DEA₄:

DEA₁=| P_b-P_a|+|P_d-P_c|

DEA₂=| P_c-P_e|+|P_e-P_b|

DEA₃=| P_c-P_a|+|P_d-P_b|

DEA₄=| P_d-P_e|+|P_e-P_a|；

1-3. is by DEA₁、DEA₂、DEA₃、DEA₄Candidate's major side direction corresponding to middle minima is as the main limit of this CTU Edge direction DE, DE i.e. represent 0 °, 45 °, 90 °, a direction in 135 °；Conforming minima is defined as follows:

DEA_min=min{DEA₁, DEA₂, DEA₃, DEA₄}。

The HEVC point self-adapted compensation method of quick sample product based on major side direction the most according to claim 1, its feature It is that step 2 detailed process is as follows:

It is the difference sum between original pixels and reconstructed pixel that 2-1. utilizes statistical information to calculate initial compensation value m=E/N, E, N is the pixel number belonging to this kind, m rounds the offset m ' after obtaining m ' and rounding and is limited between [-7,7]；

2-2. limits adjustment according to boundary compensation kind further to offset, and for kind 1 and kind 2, offset must Must be more than or equal to 0, the offset of kind 3 and kind 4 is necessarily less than equal to 0；

2-3. travels through all of candidate's offset [0, m '] or [m ', 0], chooses the offset conduct that relative rate distortion cost is minimum Optimal compensation value M, now number of coded bits R=| M |+1；Rate distortion costs Cost is defined as

Cost=Distortion+ λ × Bitrate

Original image and the distortion of reconstruct image under wherein Distortion is a certain EO pattern, λ is Lagrange factor, Bitrate is the bit number needed for coding SAO parameter information.

The HEVC point self-adapted compensation method of quick sample product based on major side direction the most according to claim 1, its feature It is the relative rate distortion cost calculating other SAO compensation ways described in step 3, the wherein determination process of BO pattern:

First, it is followed successively by 0～31 sideband and chooses optimal compensation value；

Then, it is determined that continuous 4 sidebands of optimum, the determination process of single sideband offset is as follows: utilize statistical information to calculate Initial compensation value m1=E/N1, E are the difference sums between original pixels and reconstructed pixel, and N1 is the pixel belonging to this sideband Number, rounds m1 and obtains m1 ', and the offset m1 ' after rounding is limited between [-7,7]；Obtain the m1 ' value after amplitude limit；

Subsequently, determine that candidate's offset scope is for [0, m1 '] or [m1 ', 0] according to its symbol；

Finally, travel through all candidate's offsets, choose the minimum offset of relative rate distortion cost as optimal compensation value M1, this Time number of coded bits R=| M1 |+2；Choose so that 4 continuous sidebands that rate distortion costs is minimum are the BO pattern of optimum, 4 The offset sum that sideband is corresponding is optimal compensation value；

In the rate distortion computation formula of Parameter fusion pattern, Distortion by relative rate distortion generation of adjacent block of employing parameter Valency, Bitrate is flag bit number of coded bits.