CN102595140B - Intra-frame prediction video coding method based on image inpainting and vector prediction operators - Google Patents

Abstract

The invention relates to an intra-frame prediction video coding method based on image inpainting and vector prediction operators. The method is technically characterized by comprising the following steps of: (1) calculating the rate distortion cost value RD1 of a current block based on a traditional intra-frame prediction mode in high efficiency video coding (HEVC); (2) calculating the rate distortion cost value RD2 of the current block based on an intra-frame prediction mode of a Laplace equation image inpainting method, and if the RD2 is smaller than the RD1, calculating the rate distortion cost value RD2 of the current block based on an intra-frame prediction mode of a total variation model image inpainting method; (3) calculating the rate distortion cost value RD3 of the current block based on an intra-frame prediction mode of the vector prediction operators; and (4) calculating the prediction pixel value of the current block according to the comparison result of the RD1, the RD2 and the RD3 by an encoder end, and predicting, compressing and coding the current block. The method is reasonable in design; the prediction accuracy of the conventional intra-frame prediction mode based on image inpainting is improved; and the coding rate can be reduced under the condition that the quality of the coded and decoded video is almost not changed, so that the compression efficiency of video coding is improved.

Description

Based on the intra-prediction video encoding method of image repair and vector prediction operator

Technical field

The present invention relates to video compression technology field, particularly a kind of intra-prediction video encoding method based on image repair and vector prediction operator.

Background technology

In actual video sequence, spatial coherence is an important sources of data redundancy.Aspect removal spatial redundancy, no matter in theory still in actual applications, infra-frame prediction is all simple, an effective and reliable method.Therefore, infra-frame prediction always is a focus in video coding technique research, significant for the spatial redundancy in reduction video compression data.

H.264 be to have introduced intra prediction mode than an important improvement of coding standard before, this makes it can utilize the spatial coherence of video further to improve compression efficiency.Each pixel in current to be predicted can be predicted by 17 immediate previous weighted sums of encoded pixels, i.e. 17 pixels in the upper left corner of this pixel place piece.Utilize intraframe prediction algorithm, aspect the spatial redundancies of removing between adjacent block, obtain more efficiently compression effectiveness.In H.264, for the prediction piece of 4x4,8x8,16x16 size, have respectively 9,9,4 kind of predictive mode predict.As next generation protocol H.264/AVC, high efficiency video coding (High Efficiency Video Coding, HEVC) is paid close attention to high definition/ultra high-definition video is effectively compressed and encoded.HEVC has inherited hybrid coding pattern H.264, but in modules, all adopts more advanced algorithm, to improve compression efficiency.Aspect infra-frame prediction, HEVC has adopted more efficient quaternary tree block structure, and this is corresponding to the macro block in H.264 and the concept of sub-block.In addition, compared to previous standard, HEVC has increased available predictive mode greatly, reaches as high as 34 kinds.This shows, in traditional intra-frame prediction method, in order to improve the accuracy of infra-frame prediction, mainly pattern quantity and the more complicated macroblock partitions scheme of employing by increasing infra-frame prediction, but this method has increased the data volume for representing macroblock partitions result and each sub-block prediction pattern, simultaneously, traditional infra-frame prediction adopts fixing prediction direction, cannot process adaptively real image.

The method addressing the above problem is the infra-frame prediction adopting based on image repair, and meanwhile, because infra-frame prediction can be counted as an image repair problem, therefore, the infra-frame prediction based on image repair is a current hot technology.Image repair (Image Inpainting) refers to the marginal information of utilizing area to be repaired, adopt mechanism of transmission by Information Communication in region to be repaired.It is fine that texture synthesizes aspect the large region of containing a large amount of details in reparation effect, and normally used is the method for template matches; Image repair based on solving partial differential equation is more suitable in smooth zonule; The people such as T.K Tan have proposed a kind of intra-frame prediction method and a kind of intra-frame prediction method based on priority template matches based on template matches; The image repair method based on solving partial differential equation is applied to respectively image and video compression by the people such as V.Bastani and D.Liu; The image repair method that the people such as D.Doshkove combine template matches and partial differential equation is applied in the middle of infra-frame prediction, and obtains good experimental result.But there are following two problems in the above-mentioned infra-frame prediction based on image repair: 1, the conventional infra-frame prediction based on Laplce's partial differential equation image repair cannot be processed the non-flat forms region in image effectively, especially contains the situation at edge; 2, the infra-frame prediction based on the reparation of texture composograph can effectively utilize the non local correlation of infra-frame prediction, but template matches need to be carried out a large amount of calculating, has increased the computational complexity of encoding and decoding, and real-time application is had to impact.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, a kind of intra-prediction video encoding method based on image repair and vector prediction operator of the compression efficiency that can effectively improve forecasting accuracy and Video coding is provided.

The present invention solves its technical problem and takes following technical scheme to realize:

Based on an intra-prediction video encoding method for image repair and vector prediction operator, comprise the following steps:

(1) calculate the rate distortion costs value RD1 of current block based on conventional frame inner estimation mode in HEVC;

(2) the rate distortion costs value RD2 of the intra prediction mode of calculating current block based on Laplace's equation image repair method, if rate distortion costs value RD2 is less than RD1, calculate the rate distortion costs value RD2 of the intra prediction mode of current block based on full Variation Model image repair method;

(3) the rate distortion costs value RD3 of the intra prediction mode of calculating current block based on vector prediction operator;

(4) coding side is according to the comparative result of rate distortion costs value RD1, RD2 and RD3, choose minimum rate distortion costs and be worth corresponding intra prediction mode, calculate the predicted pixel values of current block and current block is predicted, compressed and encodes by corresponding intra prediction mode.

And described rate distortion costs value RD1, rate distortion costs value RD2, rate distortion costs value RD3 is all that the rate distortion costs computing module being carried by HEVC calculates.

And the predicted pixel values that the described intra prediction mode based on full Variation Model image repair calculates current block is to obtain after completing by following iterative formula iteration:

$u_O^{\left(n\right)}=\underset{P\∈{\mathrm{\Λ}}_O}{\mathrm{\Σ}}{h}_{\mathrm{OP}}^{(n-1)}{u}_P^{(n-1)}+h_{\mathrm{OO}}^{(n-1)}{u}_O^{(n-1)}$

Wherein, h ^(n-1)=h (u ^(n-1)), n-1 represents the analog value of the parameters in last iteration process, n represents the analog value of parameters in current iteration process.

And the method for the described predicted pixel values by the intra prediction mode calculating current block based on vector prediction operator comprises the following steps:

(1) current block is divided into and equates sub-block P _i, i=1,2,3,4, i is initialized as 1, and region of search S is set to current block adjacent left-hand and top, 2x current block size and does not comprise the pixel region of current block;

(2) to current sub-block P _icarry out 1/2 down-sampling with region of search S, obtain the DP after down-sampling _iand DS;

(3) calculate DP _iminimum sad value in DS, obtains position coordinates (x, y), if sad value is more than or equal to threshold value Thres, thinks and finishes inapplicable pattern of current block to return, otherwise continue;

(4) calculate current sub-block P _iminimum sad value place coordinate in S in (2x, 2y) and adjacent position thereof, obtains current sub-block P _ipredicted value and match block vector V i, and by P _ibe incorporated to region of search S, i is from adding 1, and wherein, adjacent position is defined as (2x-1,2y), (2x+1,2y), (2x, 2y-1) and (2x, 2y+1), the piece of finding sad value minimum in above-mentioned five match block is as blocks and optimal matching blocks, if (2x, 2y), near the edge of region of search, only chooses available match block and compares;

(5) if i is less than or equal to 4, execution step (2), otherwise processing finishes and obtains predicted pixel values and the vector data of current block.

And described calculating sad value adopts following formula:

$\mathrm{SAD}=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}|S_{t\arg \mathrm{et}}\left(i\right)-S_{\mathrm{template}}\left(i\right)|$

In formula, N is the quantity of pixel in sub-block, i represent pixel position, S _targetthe block of pixels in region of search, S _templateit is selected template block of pixels.

Advantage of the present invention and good effect are:

The present invention is reasonable in design, for the intra-prediction video encoding method in HEVC, introduce based on Laplace's equation and the intra prediction mode of the enhanced image reparation that Variation Model combines entirely and the intra prediction mode based on vector prediction operator, choosing optimum pattern by comparing rate distortion cost predicts and encodes, improve the population rate distortion performance of HEVC Video Codec by the non local correlation of I picture, the method not only can solve predictive mode in traditional infra-frame prediction and fix, problem that cannot adaptive video content, improve the forecasting accuracy of the existing intra prediction mode based on image repair, and can be in the situation that encoding and decoding rear video quality be substantially constant, reduce encoder bit rate, thereby improve the compression efficiency of Video coding.

Accompanying drawing explanation

Fig. 1 is the infra-frame prediction schematic diagram based on image repair algorithm;

Fig. 2 is that full Variation Model quantizes and solves middle related pixel position definition;

Fig. 3 is macroblock partitions and region of search schematic diagram in the infra-frame prediction based on vector prediction operator.

Embodiment

Below in conjunction with accompanying drawing, the embodiment of the present invention is further described:

A kind of intra-prediction video encoding method based on image repair and vector prediction operator, as shown in Figure 1, adopt based on Laplace's equation and the intra prediction mode of the enhanced image reparation that Variation Model combines entirely and the intra prediction mode based on vector prediction operator, choose the pattern with minimum rate distortion costs by comparing rate distortion cost and current macro predicted and compressed, specifically comprise the following steps:

Step 1: calculate the rate distortion costs value RD1 of current block based on conventional frame inner estimation mode in HEVC.

In HEVC, the selection of intra prediction mode is to realize according to the calculating of rate distortion costs, selects to have the pattern of minimum rate distortion costs as the predictive mode of current macro, and therefore, HEVC has carried rate distortion costs computing module.The rate distortion costs of the intra prediction mode in HEVC and the intra prediction mode based on Laplace's equation image repair method proposed by the invention and the intra prediction mode based on vector prediction operator is all to calculate by this module, realization of the present invention does not need to be concerned about the occurrence of this numerical value, only be designated as RD1 for the convenience of explaining, lower same.

Step 2: the rate distortion costs value RD2 that calculates the intra prediction mode of current block based on Laplace's equation image repair method, if rate distortion costs value RD2 is less than RD1, calculate the rate distortion costs value RD2 of the intra prediction mode of current block based on full Variation Model image repair method.

This step is by the intra prediction mode of the enhanced image reparation forming that combines based on Laplace's equation and full Variation Model, to improve the accuracy of infra-frame prediction and the compression efficiency of infra-frame prediction, solve the problem that predictive mode in traditional infra-frame prediction is fixed, cannot adaptive video content.

Step 3: the rate distortion costs value RD3 that calculates the intra prediction mode of current block based on vector prediction operator.

Step 4: coding side is according to the comparative result of rate distortion costs value RD1, RD2 and RD3, choose minimum rate distortion costs and be worth corresponding intra prediction mode, calculate the predicted pixel values of current block and current block is predicted, compressed and encodes by corresponding intra prediction mode.

In this step, choose corresponding intra prediction mode according to the comparative result of rate distortion costs value RD1, RD2 and RD3, if i.e.: RD1 minimum, choose the predicted pixel values that intra prediction mode traditional in HEVC calculates current block, if RD2 minimum, choose the predicted pixel values of calculating current block based on the intra prediction mode of full Variation Model image repair, if RD3 minimum is chosen the predicted pixel values of calculating current block based on the intra prediction mode of vector prediction operator.Illustrate respectively that three kinds of intra prediction modes calculate the method for the predicted pixel values of current block below:

1, calculate the method for the predicted pixel values of current block by intra prediction mode traditional in HEVC:

The predicted pixel values of calculating current block by intra prediction mode traditional in HEVC, the method is to belong to the function that HEVC self has, and is not specifically described at this.

2, the intra prediction mode by based on full Variation Model image repair calculates the method for the predicted pixel values of current block and is:

The method that intra prediction mode based on Laplace's equation calculates the predicted pixel values of current block is:

In two-dimensional discrete situation, the numerical value form of Laplacian can be expressed as:

$\mathrm{Lf}(x,y)=\underset{(i,j)\∈u_4(x,y)}{\mathrm{\Σ}}f(i,j)-4f(x,y)---\left(1\right)$

Wherein, u ₄(x, y) represented 4 neighbor pixels of current pixel point f (x, y).But in the time that f (x, y) is positioned on the border of right side and below, some neighbor pixel is unavailable, must special processing.Obtain thus expression formula below:

$\mathrm{Lf}(x,y)=\underset{(i,j)\∈{\mathrm{\μ}}_4(x,y)}{\mathrm{\Σ}}\mathrm{\χ}(i,j)f(i,j)-N_{\mathrm{xy}}f(x,y)---\left(2\right)$

In the time that corresponding neighbor is available, χ (x, y) is 1, otherwise is 0, N _xyrepresent available neighbor number.In this way, the image repair algorithm based on Laplace's equation can correctly be processed the disabled situation of boundary pixel.Utilize the method for iteration to solve, its process can be expressed as follows:

f ^(t+1)＝f ^(t)+c ^(t)(Lf) ^(t)，t＝0，1，2，... (3)

Wherein, c ^(t)represent the step-length of iteration, when || f ^(τ+1)-f ^(τ)|| while being less than certain threshold value, function f ^(τ)just be considered to the predicted pixel values obtaining based on the infra-frame prediction of Laplace's equation.

The method that intra prediction mode based on full Variation Model image repair calculates the predicted pixel values of current block is:

As shown in Figure 2, for given object pixel 0, definition E, N, W and S represent respectively its four adjacent pixels, and e, n, w and s represent respectively the point on corresponding four centre positions, and these four points are imaginary points, in image, can not directly obtain.Definition:

Λ _o＝{E，N，W，S} (4)

$v=(v^1,v^2)=\▿u/|\▿u|---\left(5\right)$

Wherein,

the gradient of representative function u,

the mould of representative function u gradient, v represents the gradient of normalized function u, is a vector.Next, the difference value at centering place:

$\▿\·v=\frac{{\∂v}^1}{\∂x}+\frac{{\∂v}^2}{\∂y}\≈\frac{v_e^1-v_w^1}{h}+\frac{v_n^2-v_s^2}{h}---\left(6\right)$

Wherein, h has represented the size of grid, and often gets and do 1 in image is processed,

with

represent respectively the value of imaginary intermediate pixel e, n, w and the s superior function v of center pixel right side, left side, upside and downside.Next, further obtain the estimated value of intermediate pixel superior function v, the information at these some places is not directly to obtain from image.For example, by intermediate pixel e as an example:

${\mathrm{<figure-callout\; id="1"\; label="v\; e"\; filenames="HDA0000141865010000013.png"\; state="\{\{state\}\}">v</figure-callout>}}_e^{}=\frac{1}{\left|{\&dtri;u}_e\right|}{\left[\frac{\&PartialD;u}{\&PartialD;x}\right]}_e\&ap;\frac{1}{\left|{\&dtri;u}_e\right|}\frac{u_E-u_O}{h}---\left(7\right)$

$\left|{\&dtri;u}_e\right|\&ap;\frac{1}{h}\sqrt{{(u_E-u_O)}^2+{[(u_{\mathrm{NE}}+u_N-u_S-u_{\mathrm{SE}})/4]}^2}---\left(8\right)$

Wherein, u is pixel value function, u _ethe value that represents the u on the E of position, the rest may be inferred by analogy for it.The above-mentioned computational methods to intermediate pixel e superior function v are also applicable to intermediate pixel n, w and the s in other three directions.

Now, the discretization expression formula of Variation Model algorithm entirely:

$0=\underset{P\&Element;{\mathrm{\&Lambda;}}_{\mathrm{<figure-callout\; id="1"\; label="O"\; filenames="HDA0000141865010000013.png"\; state="\{\{state\}\}">O</figure-callout>}}}{\mathrm{\&Sigma;}}\frac{1}{\left|{\&dtri;u}_p\right|}(u_O-u_P)+{\mathrm{\&lambda;}}_e\left(O\right)(u_O-u_O^0)---\left(9\right)$

Wherein, 0 represents the arbitrary pixel in the E ∩ Ω of region, Λ _orepresent pixel set in pixel 0 four proximal direction N, S, W, S}, if P=E, p represents e so, the rest may be inferred by analogy for it, u _orepresent pixel value to be asked, function f correspondence,

represent known pixel value, function f* correspondence.Definition:

$w_P=\frac{1}{\left|{\&dtri;u}_P\right|},P\&Element;{\mathrm{\&Lambda;}}_O---\left(10\right)$

$h_{\mathrm{OP}}=\frac{w_P}{{\mathrm{\&Sigma;}}_{P\&Element;{\mathrm{\&Lambda;}}_O}{w}_P+{\mathrm{\&lambda;}}_e\left(O\right)}---\left(11\right)$

$h_{\mathrm{OO}}=\frac{{\mathrm{\&lambda;}}_e\left(O\right)}{{\mathrm{\&Sigma;}}_{P\&Element;{\mathrm{\&Lambda;}}_O}{w}_P+{\mathrm{\&lambda;}}_e\left(O\right)}---\left(12\right)$

The iterative formula finally obtaining is:

$u_O^{\left(n\right)}=\underset{P\&Element;{\mathrm{\&Lambda;}}_O}{\mathrm{\&Sigma;}}{h}_{\mathrm{OP}}^{(n-1)}{u}_P^{(n-1)}+h_{\mathrm{OO}}^{(n-1)}{u}_O^{(n-1)}---\left(13\right)$

Wherein, h ^(n-1)=h (u ^(n-1)), n-1 represents the analog value of the parameters in last iteration process, n represents the analog value of parameters in current iteration process.After iteration completes, obtain the predicted pixel values of the intra prediction mode of current block based on full Variation Model image repair.

3, the intra prediction mode by based on vector prediction operator calculates the method for the predicted pixel values of current block and is:

The core of the method is in reconstructed pixel region, to utilize near pixel value current block to obtain the predicted pixel values of current block by template matches, to utilize the non local correlation of I picture.For achieving the above object, when current macro is encoded, adopt following steps to realize:

(1) current block is divided into and equates sub-block P _i, i=1,2,3,4, i is initialized as 1, and region of search S is set to current block adjacent left-hand and top, 2x current block size and does not comprise the pixel region of current block, as shown in Figure 3.

(2) current sub-block Pi and region of search S are carried out to 1/2 down-sampling, obtain DPi and DS after down-sampling.

(3) calculate DP _iminimum sad value in DS, obtains position coordinates (x, y), if sad value is more than or equal to threshold value Thres, thinks and finishes inapplicable pattern of current block to return, otherwise continue.Wherein, the calculation expression of sad value is:

$\mathrm{SAD}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}|S_{t\arg \mathrm{et}}\left(i\right)-S_{\mathrm{template}}\left(i\right)|---\left(14\right)$

Wherein, N is the quantity of pixel in sub-block, i represent pixel position, S _targetthe block of pixels in region of search, S _templateit is selected template block of pixels.

(4) calculate Pi minimum sad value place coordinate in (2x, 2y) and adjacent position thereof in S, obtain predicted value and the match block vector V i of current sub-block Pi, and Pi is incorporated to region of search S, oneself adds 1 i.Wherein, adjacent position is defined as (2x-1,2y), (2x+1,2y), (2x, 2y-1) and (2x, 2y+1).The piece of finding sad value minimum in above-mentioned five match block is as blocks and optimal matching blocks.In addition, if the edge of (2x, 2y) close region of search, not all four adjacent match block all can be used, and now, only choose available match block and compare.

(5) if i is less than or equal to 4, execution step (2), otherwise finish to return, obtain predicted pixel values and the vector data of current block.

The simulated environment adopting in case verification is HEVC reference software HM 2.0, and the intra prediction mode based on enhanced image reparation algorithm and vector prediction operator proposed by the invention has been embedded in the intra-framed prediction module of HM2.0.Specifically, this intra prediction mode is to have added in the model selection process of coding unit CU as an additional modes, and the standard of model selection is that the rate distortion costs function in the HEVC of establishing criteria calculates.If the rate distortion costs functional value of new predictive mode is less than in HEVC self predictive mode, will use new predictive mode proposed by the invention so, otherwise, will use in HEVC the predictive mode of self.In order to assess the performance of new predictive mode in the present invention, 5 video sequences are used altogether.These 5 video sequences are chosen the video sequence collection adopting when motion is assessed to technology from JCT-VC, and resolution from low to high.The setting parameter while testing below: in frame, the maximum TU degree of depth is that 3, TU minimum value is 4, and maximum is 32; CABAC entropy coding and RDOQ open; In addition, each video sequence uses full frame internal schema to encode, therefore Intra Period and GOP Size are 1.Finally being used for assessing the PSNR value (Y-PSNR) of compression performance and statistical tool that encoder bit rate is carried by HM2.0 completes.

Table 1 has provided intra prediction mode encoder bit rate case verification result table proposed by the invention, for the ease of contrast, has provided the encoder bit rate result of repairing based on vector prediction, Laplace's equation and enhanced image simultaneously.All bit rate variation are all compared with original intra prediction mode coding result in HM2.0.The coded frame quantity of each video sequence is 10, and quantization parameter QP is set to 32.

Table 1

Table 2 has provided intra prediction mode PSNR case verification result table proposed by the invention, and the coded frame of each video sequence is 10 frames, and quantization parameter QP is 32.As can be seen from Table 2, compared with intraframe predictive coding pattern original in HM2.0, PSNR remains unchanged substantially.

Table 2

Associative list 1 and table 2, we can find out, the compression performance of repairing the intra prediction mode of algorithm based on enhanced image is clearly better than the intra prediction mode based on Laplace's equation, when enhanced image is repaired to algorithm and vector prediction operator and is combined, the highest 4.51% the bit rate of can reaching is saved, and average result is 1.65%.Therefore, case verification result shows, new predictive mode proposed by the invention, in the situation that PSNR remains unchanged substantially, has reached average 1.65% bit rate saving.

It is emphasized that; embodiment of the present invention is illustrative; rather than determinate; therefore the present invention is not limited to the embodiment described in embodiment; every other execution modes that drawn by those skilled in the art's technical scheme according to the present invention, belong to the scope of protection of the invention equally.

Claims (5)

1. the intra-prediction video encoding method based on image repair and vector prediction operator, is characterized in that: comprise the following steps:

(1) calculate the rate distortion costs value RD1 of current block based on conventional frame inner estimation mode in HEVC;

(2) the rate distortion costs value RD2 of the intra prediction mode of calculating current block based on Laplace's equation image repair method, if rate distortion costs value RD2 is less than RD1, calculate the rate distortion costs value RD2 of the intra prediction mode of current block based on full Variation Model image repair method;

(3) the rate distortion costs value RD3 of the intra prediction mode of calculating current block based on vector prediction operator;

(4) coding side is according to the comparative result of rate distortion costs value RD1, RD2 and RD3, choose minimum rate distortion costs and be worth corresponding intra prediction mode, calculate the predicted pixel values of current block and current block is predicted, compressed and encodes by corresponding intra prediction mode.

2. the intra-prediction video encoding method based on image repair and vector prediction operator according to claim 1, is characterized in that: described rate distortion costs value RD1, rate distortion costs value RD2, rate distortion costs value RD3 is all that the rate distortion costs computing module being carried by HEVC calculates.

3. the intra-prediction video encoding method based on image repair and vector prediction operator according to claim 1, is characterized in that: the predicted pixel values that the described intra prediction mode based on full Variation Model image repair calculates current block is to obtain after completing by following iterative formula iteration:

$u_O^{\left(n\right)}=\underset{P\&Element;{\mathrm{\&Lambda;}}_O}{\mathrm{\&Sigma;}}{h}_{\mathrm{OP}}^{(n-1)}{u}_P^{(n-1)}+h_{\mathrm{OO}}^{(n-1)}{u}_O^{(n-1)}$

Wherein, h ^(n-1)=h (u ^(n-1)), n-1 represents the analog value of the parameters in last iteration process, n represents the analog value of parameters in current iteration process.

4. the intra-prediction video encoding method based on image repair and vector prediction operator according to claim 1, is characterized in that: the method for the described predicted pixel values by the intra prediction mode calculating current block based on vector prediction operator comprises the following steps:

(1) current block is divided into and equates sub-block P _i, i=1,2,3,4, i is initialized as 1, and region of search S is set to current block adjacent left-hand and top, 2x current block size and does not comprise the pixel region of current block;

(2) to current sub-block P _icarry out 1/2 down-sampling with region of search S, obtain the DP after down-sampling _iand DS;

(3) calculate DP _iminimum sad value in DS, obtains position coordinates (x, y), if sad value is more than or equal to threshold value Thres, thinks and finishes inapplicable pattern of current block to return, otherwise continue;

(4) calculate current sub-block P _iminimum sad value place coordinate in S in (2x, 2y) and adjacent position thereof, obtains current sub-block P _ipredicted value and match block vector V i, and by P _ibe incorporated to region of search S, i is from adding 1, and wherein, adjacent position is defined as (2x-1,2y), (2x+1,2y), (2x, 2y-1) and (2x, 2y+1), the piece of finding sad value minimum in above-mentioned five match block is as blocks and optimal matching blocks, if (2x, 2y), near the edge of region of search, only chooses available match block and compares;

(5) if i is less than or equal to 4, execution step (2), otherwise processing finishes and obtains predicted pixel values and the vector data of current block.

5. the intra-prediction video encoding method based on image repair and vector prediction operator according to claim 4, is characterized in that: described calculating sad value adopts following formula:

$\mathrm{SAD}=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}|S_{t\arg \mathrm{et}}\left(i\right)-S_{\mathrm{template}}\left(i\right)|$

In formula, N is the quantity of pixel in sub-block, i represent pixel position, S _targetthe block of pixels in region of search, S _templateit is selected template block of pixels.

Images