CN101163249B - DC mode prediction technique - Google Patents

Abstract

The invention relates to a direct current mode predicting method which includes selecting current prediction macro block and dividing the macro block into a plurality of image element blocks; a image element block is selected as the current image element block according to an image point, which is placed at the same row of the image element point to be predicted in the current image element block in the left image element block of the current image element block and the image element allays next to the current image element block; and/or the image element point which is in the upper part of the current image element and in the image element row of the image element block next to the current image element block and is placed at the same row of the image element point in the current image element block to be predicted; and/or in the top-left image element block of the current image element block, and the image element point which is next to the current image element block, predicting the image element value of the image element point to be predicated in the current image element block. The invention implements filtering and weighted process on the coded and available image element point, which is not simple equaling process; the weighted value after filtering is used as the prediction value of current image element, which improves the accuracy of prediction.

Description

DC mode prediction technique

Technical field

The present invention relates to a kind of intra-frame prediction method, especially a kind of DC mode prediction technique.

Background technology

Existing video encoding standard all is to be based upon on the mixed coding framework of transition coding and motion compensation, reduces the spatial coherence of video sequence with transition coding, reduces the relativity of time domain of video sequence with motion compensation.As shown in Figure 1, video encoding standard adopts a series of state-of-the-art technologies, comprises infra-frame prediction, inter prediction, conversion, quantification and entropy coding etc., and the employing of these technology makes video encoding standard obtain very high code efficiency.Wherein the introducing of infra-prediction techniques is one of key factor of video coding performance raising.

Infra-frame prediction usage space predictive mode comes the redundancy in the removal of images, and it can independent encoding and decoding, improve accuracy of predicting and quality, as the reference frame of subsequent sequence, for video playback provides random access point.In existing infra-prediction techniques, the prediction of the pixel value of brightness or colourity adopt based on p * q piece predict, wherein p represents the columns of piece, q represents the line number of piece, it be utilize on this p * q piece, upper right, left, upper left and lower-left reconstructed pixels value, predict the pixel value of the pixel of current pixel piece according to some predictive mode and computation rule thereof, as shown in Figure 2, the row-coordinate of i remarked pixel point, value is i=0,1 ..., 2q; The row coordinate of j remarked pixel point, value are j=0,1 ..., 2p; PredMatrix[i, j] represent that the row-coordinate in the current pixel piece is i, the row coordinate is the pixel value of the pixel of j.If C[i] (i=0,1, ..., 2 q) pixel of the capable pixel of i of row adjacent in the expression current pixel piece left pixel piece with this current pixel piece, R[j] (j=0,1 ..., 2p) pixel value of the j row pixel of the delegation adjacent, wherein C[0 in the expression current pixel piece upside block of pixels with the current pixel piece]=R[0].To any given coordinate is (i, j) pixel, the pixel C[i of the reconstruct of having encoded around utilizing] (i=0,1, ..., 2q) and R[j] (j=0,1 ..., 2p) predict, according to different standards and different prediction block sizes, different predictive modes is arranged, the following describes AVS standard and the predictive mode in the standard H.264/AVC.

In video coding, all be to be unit with 16 * 16 macro block to the processing of video image.In the AVS standard, a macro block comprises 48 * 8 luminance block and 28 * 8 chrominance block.Luminance block has 5 kinds of predictive modes: vertical prediction pattern, horizontal forecast pattern, direct current (DirectCurrent is hereinafter to be referred as DC) predictive mode, diagonal angle bottom left predictive mode, diagonal angle bottom right predictive mode.Except that the DC predictive mode, other all be directional prediction modes, Fig. 3 shows the sensing of directional prediction modes, and wherein, 0,1,3,4 represent the sensing of vertical prediction pattern, the sensing of horizontal pattern prediction, the sensing of diagonal angle bottom left predictive mode, the sensing of diagonal angle bottom right predictive mode respectively; C represents the left side one row adjacent with the current pixel piece, and r represents the top delegation adjacent with the current pixel piece.

Unshowned DC predictive mode is defined as follows:

1) if r[i], c[i] (i=0～9) " available " (known) all, if the piece at certain sample place does not exist or this sample is not decoded, this sample " unavailable " then, otherwise, this sample " available ", then

predMatrix[x，y]＝((r[x]+2×c[x+1]+c[x+2]+2)＞＞2+(c[y]+2×c[y+1]+c[y+2]+2)＞＞2)＞＞1(x，y＝0～7)

Wherein, predMatrix[x, y] represent that coordinate is (x, the pixel value of pixel to be predicted y) in the current pixel piece; R[x+1] in the expression top delegation adjacent, be positioned at the pixel value of the pixel of same row with pixel to be predicted with the current pixel piece; R[x] expression r[x+1] pixel value of left side neighbor pixel of pixel of expression; R[x+2] expression r[x+1] pixel value of right side neighbor pixel of pixel of expression; C[x+1] the expression left side adjacent with the current pixel piece one be in being listed as, and is positioned at pixel value with the pixel of delegation with pixel to be predicted; C[x] expression c[x+1] pixel value of upside pixel of pixel of expression; C[x+2] expression c[x+1] pixel value of downside pixel of pixel of expression.In the following analysis, above-mentioned symbol is the implication of above-mentioned statement.

2) if having only r[i] (i=0～9) " available ", then

predMatrix[x，y]＝(c[y]+2×c[y+1]+c[y+2]+2)＞＞2(x，y＝0～7)

3) if having only c[i] (i=0～9) " available ", then

predMatrix[x，y]＝(c[y]+2×c[y+1]+c[y+2]+2)＞＞2(x，y＝0～7)；

4) otherwise, predMatrix[x, y]=128 (x, y=0～7)

Chrominance block has 4 kinds of patterns:

Pattern 0:DC predicts (DC prediction)

Pattern 1: horizontal forecast (horizontal prediction)

Pattern 2: vertical prediction (vertical prediction)

Mode 3: dull and stereotyped prediction (plane prediction)

Wherein the definition of DC predictive mode is consistent with the definition of luminance block 8 * 8 block prediction modes.

In standard H.264/AVC, a macro block comprises a luminance block and two chrominance block, and luminance block is 1 16 * 16 luminance block or 16 4 * 4 luminance block, and chrominance block is 28 * 8.Luminance block for 4 * 4 has 9 kinds of predictive modes, and these predictive modes and order are:

Pattern 0: vertical prediction (vertical prediction)

Pattern 1: horizontal forecast (horizontal prediction)

Pattern 2:DC predicts (DC prediction)

Mode 3: 45 degree direction predictions (diagonal down/left prediction)

Pattern 4:135 degree direction prediction (diagonal down/right prediction)

Pattern 5:112.5 degree direction prediction (vertical-right prediction)

Pattern 6:157.5 degree direction prediction (horizontal-down prediction)

Mode 7: 67.5 degree direction predictions (vertical-left prediction)

Pattern 8:22.5 degree direction prediction (horizontal-up prediction)

Wherein, except the DC predictive mode, remaining 8 kinds of predictive modes are called directional prediction modes, and Fig. 4 shows reference pixel and prediction piece, and what wherein capitalization was represented is reference pixel, and what lowercase was represented is the prediction piece.Numeral among Fig. 5 has been indicated the sensing of all directions predictive modes.

The DC predictive mode that does not mark is defined as:

1) if r[i], c[i] (i=1,2 ..., 4) all available, predict that then the pixel predictors of all forecast samples in the piece is:

$\mathrm{predMatrix}[x,y]=(\underset{x=1}{\overset{4}{\mathrm{\Σ}}}r[x]+\underset{y=1}{\overset{4}{\mathrm{\Σ}}}c[y]+4)/8(x,y=0~3);$

2) if having only r[i] (i=1,2 ..., 4) available, then

$\mathrm{predMatrix}[x,y]=(\underset{x=1}{\overset{4}{\mathrm{\Σ}}}r\left[x\right]+2)/4(x,y=0~3);$

3) if having only c[i] (i=1,2 ..., 4) available, then

$\mathrm{predMatrix}[x,y]=(\underset{y=1}{\overset{4}{\mathrm{\Σ}}}c[y]+2)/4(x,y=0~3);$

4) otherwise, predMatrix[x, y]=128 (x, y=0～3)

16 * 16 luminance block have 4 kinds of patterns, are respectively:

Pattern 0: vertical prediction (vertical prediction)

Pattern 1: horizontal forecast (horizontal prediction)

Pattern 2:DC predicts (DC pre diction)

Mode 3: dull and stereotyped prediction (plane prediction)

The definition of wherein DC predictive mode and luminance block are that 4 * 4 predictive mode defines consistent.

8 * 8 chrominance block have 4 kinds of patterns, are respectively:

Pattern 0:DC predicts (DC prediction)

Pattern 1: horizontal forecast (horizontal prediction)

Pattern 2: vertical prediction (vertical prediction)

Mode 3: dull and stereotyped prediction (plane prediction)

The definition of wherein DC predictive mode and luminance block are that 4 * 4 predictive mode defines consistent.

From above-mentioned analysis as can be seen: the AVS standard, be by to coordinate for (x, the available pixel value r[x that current pixel y) is relevant], r[x+1], r[x+2], c[x], c[x+1], c[x+2] filtered pixel average is as the predicted value of current pixel; H.264/AVC in the standard, be the predicted value of the mean value of the encoded available pixel in the left side by the current pixel piece and top as the current pixel piece.It is very coarse to calculate average method after the average or filtering of this direct calculating, can not fully disclose the correlation between pixel, and therefore, this Forecasting Methodology is coarse.

Summary of the invention

The objective of the invention is defective, a kind of DC mode prediction method is provided,, realize accurately predicting each pixel to be predicted in the current pixel piece in order to solve existing Forecasting Methodology inaccuracy problem at prior art.The present invention provides following technical scheme by embodiment:

A kind of DC mode prediction method comprises: choose the macro block of current prediction, this macro block is divided into several block of pixels; Choose a block of pixels as the current pixel piece, in the left pixel piece of this current pixel piece with this current pixel piece adjacent pixels row in, the pixel that is positioned at delegation with pixel to be predicted in this current pixel piece is first pixel, in the upside block of pixels of this current pixel piece with this current pixel piece adjacent pixels row in, the pixel that is positioned at same row with pixel to be predicted in this current pixel piece is second pixel, in the upper left side block of pixels of this current pixel piece, with this current pixel piece adjacent pixels point be the 3rd pixel; If the pixel value of all pixels in the pixel column at the pixel column at the described first pixel place and the second pixel place, and the pixel value of the 3rd pixel is all known, then described first pixel is carried out Filtering Processing and obtain first value, described second pixel is carried out Filtering Processing obtain second value, described the 3rd pixel is carried out Filtering Processing obtain the 3rd value; Described first value, second value and the 3rd value be weighted handle the pixel value obtain described pixel to be predicted; If have only the pixel value of the pixel in the pixel column at the described first pixel place known, then will carry out first value that obtains after the Filtering Processing pixel value to described first pixel as described pixel to be predicted; If have only the pixel value of the pixel in the pixel column at the described second pixel place known, then will carry out second value that obtains after the Filtering Processing pixel value to described second pixel as described pixel to be predicted; When if the pixel value of all pixels in the pixel column at the pixel column at the described first pixel place and the second pixel place is all unknown, the pixel value of arbitrary pixel in the described current pixel piece is made as 128.

The embodiment of the invention is carried out filtration combined weighted processing by the pixel value to the available pixel of encoding, rather than simple average treatment, with the filtered weighted value of these pixel values, as the predicted value of current pixel, improves accuracy of predicting.

Description of drawings

Fig. 1 is a prior art video coding combination frame schematic diagram;

Fig. 2 is the coordinate schematic diagram of the current prediction piece of prior art;

Fig. 3 is the sensing figure of direction prediction in the prior art AVS standard;

Fig. 4 is the H.264/AVC schematic diagram of reference pixel and current prediction piece relation in the standard of prior art;

Fig. 5 is the H.264/AVC sensing figure of direction prediction in the standard of prior art;

Fig. 6 is a DC mode prediction method flow chart of the present invention;

Fig. 7 is the graph of a relation of pixel in DC mode prediction method current pixel piece of the present invention and the adjacent pixel blocks.

Embodiment

Further specify technical scheme of the present invention below in conjunction with the drawings and specific embodiments.

Fig. 6 is a DC mode prediction method flow chart of the present invention, and Fig. 7 is the graph of a relation of pixel in DC mode prediction method current pixel piece of the present invention and the adjacent pixel blocks.Referring to Fig. 6 and Fig. 7, this DC mode prediction method comprises: step 11: choose the macro block of current prediction, this macro block is divided into several block of pixels; Because the processing of video image is to be that unit handles with 16 * 16 macro block, so this macro block is elected 16 * 16 pixels as.With this macro block by from left to right, from top to bottom order is divided into the block of pixels of several p * q; P represents columns, and q represents line number, and the numerical value of p can be taken as 4,8,16, and the numerical value of q can be taken as 4,8,16.Step 12: choose a block of pixels as current pixel piece 1, according in the left pixel piece 2 of this current pixel piece with these current pixel piece 1 adjacent pixels row 21 in, be positioned at first pixel with pixel to be predicted in this current pixel piece with delegation; And/or in the upside block of pixels 3 of this current pixel piece with this current pixel piece 1 adjacent pixels row 31 in, be positioned at second pixel of same row with pixel to be predicted in this current pixel piece; And/or in the upper left side block of pixels of this current pixel piece 4, three pixel adjacent with this current pixel piece predicted the pixel value of this pixel to be predicted in this current pixel piece.

Step 12 in the above-mentioned flow chart, specific embodiment can be divided into first embodiment---8 * 8 luminance block and/or 8 * 8 chrominance block and/or DC mode prediction method of 8 * 8 chrominance block under the standard H.264/AVC under the AVS standard, and second embodiment---the DC mode prediction method of 4 * 4 luminance block under the standard H.264/AVC.

First embodiment: before carrying out the DC prediction, will obtain reference sample (pixel) earlier, predict with the foundation reference sample.If the image pattern matrix under the current pixel piece is 8 * 8 matrix I, I can represent brightness or chrominance matrix.If not " existence " or this sample decoding as yet of the piece at certain sample place, this sample " unavailable " then, otherwise this sample " available ".

If the coordinate of current pixel piece upper left corner sample be (x0, y0), wherein, x0 represents row, y0 represents row, then current pixel piece reference sample obtains by following rule:

If coordinate be (promptly the row adjacent with the current pixel piece is available in the last piece of current pixel piece for x0+i-1, the y0-1) sample of (i=1..8) " available ", and then this row element (pixel value of reference sample) is: r[i]=I[y0-1] [x0+i-1]; Otherwise r[i] " unavailable ";

If coordinate be (promptly the row adjacent with the current pixel piece is available in upper right of the current pixel piece, then r[i for x0+i-1, the y0-1) sample of (i=9..16) " available "]=I[y0-1] [x0+i-1]; Otherwise r[i] equal r[8];

If coordinate be (promptly row adjacent with the current pixel piece are available in the left piece of current pixel piece, then c[i for x0-1, the y0+i-1) sample of (i=1..8) " available "]=I[y 0+i-1] [x0-1]; Otherwise c[i] " unavailable ";

If coordinate be (promptly row adjacent with the current pixel piece are available in the lower-left piece of current pixel piece, then c[i for x0-1, the y0+i-1) sample of (i=9..16) " available "]=I[y0+i-1] [x0-1]; Otherwise c[i] equal c[8];

If coordinate be (x0-1, sample y0-1) " available ", then r[0]=I[y0-1] [x0-1]; Otherwise: if r[1] " available ", and c[1] " available ", r[0]=(r[1]+c[1]+1)＞＞1;

If r[1] " available ", and c[1] " unavailable ", r[0]=r[1];

If c[1] " available ", and r[1] " unavailable ", r[0]=c[1];

Otherwise r[0] " unavailable ".

After obtaining reference sample, will carry out pixel value DC prediction to the current pixel piece, computational methods are as follows:

(1) if r[i], c[i] (i=0...9) all available, be the row adjacent in the piece of a current pixel piece left side with the current pixel piece, the row adjacent in the piece on the current pixel piece with the current pixel piece, but and in upper left of the current pixel piece with the equal time spent of current pixel piece adjacent pixels point, in row adjacent in the piece of a current pixel piece left side with the current pixel piece, be positioned at the pixel value of first pixel of going together with pixel to be predicted, on the current pixel piece in the piece in the row adjacent with the current pixel piece, be positioned at the pixel value of second pixel of same column with pixel to be predicted, the pixel value of three pixel adjacent in upper left of the current pixel piece with the current pixel piece, the pixel value of the pixel to be predicted in the prediction current pixel piece.The pixel value that is specially described first pixel carries out filtering, obtains first value, and the pixel value of described second pixel is carried out filtering, obtains second value, and the pixel value of described the 3rd pixel is carried out filtering, obtains the 3rd value; Described first value, second value and the 3rd value are weighted processing, with the pixel value of the value after the weighting as pixel to be predicted.

If the line number of current pixel piece and columns are all greater than 8 o'clock (present embodiment is this situation), pixel value to described first pixel carries out filtering, obtaining first value is specially: the weight of the pixel value of last pixel that will be adjacent with described first pixel is made as 1, the weight of the pixel value of described first pixel is made as 2, the weight of the pixel value of following pixel that will be adjacent with described first pixel is made as 1, with the mean value of the data after the above-mentioned weighted, as described first value; Pixel value to described second pixel carries out filtering, obtaining second value is specially: the weight of the pixel value of left pixel that will be adjacent with described second pixel is made as 1, the weight of the pixel value of described second pixel is made as 2, the weight of the pixel value of right pixel that will be adjacent with described second pixel is made as 1, with the mean value of the data after the above-mentioned weighted, as described second value; Pixel value to described the 3rd pixel carries out filtering, obtaining the 3rd value is specially: the weight of the pixel value of right pixel that will be adjacent with described the 3rd pixel is made as 1, the weight of the pixel value of described the 3rd pixel is made as 2, the weight of the pixel value of following pixel that will be adjacent with described the 3rd pixel is made as 1, with the mean value of the data after the above-mentioned weighted, as described the 3rd value.

Be exemplified below: the current pixel piece with above-mentioned 8 * 8 is an example, as to want the prediction line coordinate be y 1, the row coordinate is the pixel value of the pixel of x1, and the pixel value of first pixel is c[y1+1 so], first value is EP[-(y1+1)]=(c[y1]+2 c[y1+1]+c[y1+2]+2)＞＞2; The pixel value of second pixel is r[x1+1], second value is EP[(x1+1)]=(r[x1]+2 r[x1+1]+r[x 1+2]+2)＞＞2; The pixel value of the 3rd pixel is r[0], the 3rd value is EP[0]=(r[1]+2 r[0]+c[1]+2)＞＞2; Above-mentioned first value, second value, the 3rd value are weighted processing again, and the value after the processing is as predicted value, i.e. predicted value predMatrix[x, y]=α EP[(x+1)]+β EP[-(y+1)]+γ EP[0].Wherein power weight coefficient α, β, γ obtain according to the reproducing kernel interpolating function, and the interpolation knot of reproducing kernel interpolating function is the adjacent left piece of current pixel piece, goes up piece, lower-left piece, upper right pixel.Following surface analysis reproducing kernel function:

The space that the reproducing kernel function forms is the reproducing kernel space, the reproducing kernel space is the important function family of a class, it is the ideal space framework of numerical analysis research, reproducing kernel theory combines with practical application, be widely used at aspects such as signal processing, quantum mechanics, random processes, particularly the reproducing kernel method provides a desirable framework for approaching of random polynary discrete data.Why reproducing kernel has so, and powerful numerical value expressive force should give the credit to the regeneration attribute that the reproducing kernel space is had fully.

W ₂ ^p(R) space is an inner product space that is defined in the absolutely continuous function on the R, and f ^(p)∈ L ²(R), the inner product in this space and norm are:

$(f\left(x\right),g\left(x\right))=\underset{R}{\∫}\underset{k=0}{\overset{p}{\mathrm{\Σ}}}{c}_k{f}^{\left(k\right)}\left(x\right)g^{\left(k\right)}\left(x\right)\mathrm{dx},f\left(x\right),g\left(x\right)\∈W_2^p\left(R\right)---\left(1\right)$

${\left|\right|f\left(x\right)\left|\right|}_{W_2^p\left(R\right)}={(f\left(x\right),f\left(x\right))}^{1/2},f\left(x\right)\∈W_2^p\left(R\right)---\left(2\right)$

Here { c _k} ^k _P=0Satisfy condition ${(a+b)}^p=\underset{k=0}{\overset{p}{\mathrm{\Σ}}}{c}_k{a}^k{b}^{p-k}.$

Note W ₂ ^p(R) reproducing kernel is R _p(x, ξ).According to the regeneration attribute of reproducing kernel, for arbitrarily $f\left(x\right)\∈W_2^p\left(R\right)$ Have

f(x)＝(f(ξ)，R _p(x，ξ)) (3)

For W ₂ ^p(R), along with the increase of p, the smoothness in reproducing kernel space strengthens, for p=1,2,3, the function in corresponding reproducing kernel space has analytical expression (seeing Table 1), function when corresponding reproducing kernel space, p＞4 does not have analytical expression, for the ease of characterizing the correlation between pixel, the interpolating function of structure should have following good characteristic: interpolation knot is accurately set up, and node can non-regular distribution, have succinct expression formula, interpolating function has certain slickness.

Table 1

In order to construct the interpolating function that satisfies These characteristics, present embodiment adopts reproducing kernel $W\left(D\right)={\mathrm{<figure-callout\; id="2"\; label="W"\; filenames="S2007101777258E00011.png,S2007101777258E00021.png"\; state="\{\{state\}\}">W</figure-callout>}}_2^3\left(R\right)\&CircleTimes;{\mathrm{<figure-callout\; id="2"\; label="W"\; filenames="S2007101777258E00011.png,S2007101777258E00021.png"\; state="\{\{state\}\}">W</figure-callout>}}_2^3\left(R\right)$ The interpolation form, W (D) is the reproducing kernel space W here ₂ ³(R) and W ₂ ³(R) product space.The reproducing kernel of W (D) is

$K(x,\mathrm{\&xi;},y,\mathrm{\&eta;})=\left(\frac{1}{16}({(x-\mathrm{\&xi;})}^2+3|x-\mathrm{\&xi;}|+3)e^{-|x-\mathrm{\&xi;}|}\right)\left(\frac{1}{16}({(y-\mathrm{\&eta;})}^2+3|y-\mathrm{\&eta;}|+3)e^{-|y-\mathrm{\&eta;}|}\right).$ (x, y) ∈ W (D) has for f arbitrarily

f(x，y)＝(f(ξ，η)，K(x，ξ，y，η)) (4)

Suppose { x _i, y _i} _i ^∞Be the point set of two inequalities, ψ _i(x, y)=K (x _i, x, y _i, y), i=1,2 ..., by to { ψ _i(x, y) } _I=1 ^∞Orthogonalization process can obtain the orthonormal basis { ψ of W (D) _i ^*(x, y) } _I=1 ^∞,

${\mathrm{\&psi;}}_i^{*}(x,y)=\underset{i=1}{\overset{i}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{\mathrm{ij}}{\mathrm{\&psi;}}_j(x,y),i=\mathrm{1,2},...---\left(5\right)$

β wherein _Ij, j=1,2 .., i; I=1,2 ..., n is the orthogonalization coefficient, can accurately be obtained by (6) formula

$\left\{\begin{array}{c}{\mathrm{\&beta;}}_{\mathrm{ii}}=1/\sqrt{C_i-\underset{k=1}{\overset{i-1}{\mathrm{\&Sigma;}}}{\stackrel{\&OverBar;}{C}}_{\mathrm{<figure-callout\; id="2"\; label="ik"\; filenames="S2007101777258E00011.png,S2007101777258E00021.png"\; state="\{\{state\}\}">ik</figure-callout>}}^2},k=\mathrm{1,2},...,i-1\\ {\mathrm{\&beta;}}_{\mathrm{il}}=-\underset{k=l}{\overset{i-1}{\mathrm{\&Sigma;}}}{\stackrel{\&OverBar;}{C}}_{\mathrm{ik}}{\mathrm{\&beta;}}_{\mathrm{kl}}/\sqrt{C_i-\underset{k=1}{\overset{i-1}{\mathrm{\&Sigma;}}}{\stackrel{\&OverBar;}{C}}_{\mathrm{<figure-callout\; id="2"\; label="ik"\; filenames="S2007101777258E00011.png,S2007101777258E00021.png"\; state="\{\{state\}\}">ik</figure-callout>}}^2},l=\mathrm{1,2},...,i-1;i=\mathrm{1,2},...,n\\ {\mathrm{\&beta;}}_{11}=1/\sqrt{C_1}\end{array}\right.---\left(6\right)$

Here C _i=(ψ _i(x, y), ψ _i(x, y)), ${\stackrel{\&OverBar;}{C}}_{\mathrm{ik}}=({\mathrm{\&psi;}}_k^{*}(x,y),{\mathrm{\&psi;}}_i(x,y)).$

Utilize the regeneration attribute in reproducing kernel space to construct reproducing kernel space interpolation formula:

$f_n(x,y)=\underset{l=1}{\overset{n}{\mathrm{\&Sigma;}}}{h}_l(x,y)f(x_l,y_l)---\left(7\right)$

Here $h_l(x,y)=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\mathrm{\&beta;}}_{\mathrm{il}}{\mathrm{\&psi;}}_i^{*}(x,y).$

Because the inner product operation in reproducing kernel space is accurate, and then the orthogonalization coefficient do not have rounding error, and special reproducing kernel has analytical expression, so interpolation coefficient h _lCan accurately obtain, promptly interpolation formula is accurately set up at the interpolation knot place.Do not apply any constraint owing to interpolation knot is distributed, this interpolation method is applicable to interpolation knot regular distribution and irregular distribution interpolation problem, when the adjacent block reconstructed pixels during as the interpolation knot of this interpolating function, just can be obtained the predicted value of current pixel.Can get by above-mentioned theory analysis, will replace original DC mode prediction method, can improve precision of prediction effectively based on the new DC mode prediction method of reproducing kernel interpolation structure.

For the ease of the reproducing kernel interpolation formula is applied to image processing, formula (7) is expressed as its equivalent form of value:

$\stackrel{\&OverBar;}{f}(x,y)=\underset{i=1}{\overset{m}{\mathrm{\&Sigma;}}}\underset{j=1}{\overset{n}{\mathrm{\&Sigma;}}}{H}_{i,j}(x,y)f(x_i,y_j)---\left(8\right)$

Here f (x _i, y _j), H _{I, j}(x, y) respectively with f (x _l, y _l), h _l(x, y) represent same implication, different just is expressed as two index (i by an original index l form, j) form, in infra-frame prediction, the prediction of the sample value of brightness or colourity adopt based on p * q piece predict, wherein p represents the columns of piece, q represents the line number of piece, it be utilize on this p * q piece, upper right, left and upper left limit reconstructed pixels value, therefore we can only select the reconstructed pixels point adjacent with the current pixel piece as interpolation knot, therefore following formula are rewritten as following form:

$\stackrel{\&OverBar;}{I}(x,y)=\underset{j=1}{\overset{2\mathrm{<figure-callout\; id="0"\; label="q\; H"\; filenames="S2007101777258E00021.png"\; state="\{\{state\}\}">q</figure-callout>}}{\mathrm{\&Sigma;}}}{H}_{}{}_{0,j}(x,y)I(x_0,y_j)+\underset{i=1}{\overset{2p}{\mathrm{\&Sigma;}}}{H}_{i,0}(x,y)I(x_i,y_0)+{\mathrm{<figure-callout\; id="0"\; label="H"\; filenames="S2007101777258E00021.png"\; state="\{\{state\}\}">H</figure-callout>}}_{\mathrm{0,0}}(x,y)I(x_0,y_0)---\left(9\right)$

Wherein,

The pixel predictors of expression current pixel piece; I (x ₀, y _j) pixel value in expression and the left side of current pixel piece, the lower-left piece in the row adjacent with the current pixel piece; I (x _i, y ₀) pixel value in the going up of expression and current pixel piece, upper right in the row adjacent with the current pixel piece; I (x ₀, y ₀) among upper left of expression current pixel piece with the pixel value of current pixel piece adjacent pixels point; The weights of their fronts calculate according to above-mentioned reproducing kernel interpolating function, in the practical application, because most of weights all are approximately 0, in order to reduce the complexity of DC mode prediction method, obtain formula (9) is approximate:

$\stackrel{\&OverBar;}{I}({\mathrm{<figure-callout\; id="0"\; label="x\; i"\; filenames="S2007101777258E00021.png"\; state="\{\{state\}\}">x</figure-callout>}}_i,y_{j0})=\mathrm{\&alpha;I}(x_0,y_{j0})+\mathrm{\&beta;I}({\mathrm{<figure-callout\; id="0"\; label="x\; i"\; filenames="S2007101777258E00021.png"\; state="\{\{state\}\}">x</figure-callout>}}_i,y_0)+\mathrm{\&gamma;I}(x_0,y_0)$

Wherein, the coordinate of pixel to be predicted is (x _I0, y _J0); The pixel value of pixel to be predicted is

I (x ₀, y _J0) for to being positioned at pixel value with the pixel of delegation with pixel to be predicted in the left piece of current pixel piece, promptly first pixel carries out filtered value; I (x _I0, y ₀) be that promptly second pixel carries out filtered value to be positioned at the pixel value of the pixel of same row in the last piece of current pixel piece with pixel to be predicted; I (x ₀, y ₀) be among upper left to the current pixel piece with the pixel value of current pixel piece pixel, promptly the 3rd pixel carries out filtered value.

According to the formula in above-mentioned: coordinate is (x, the pixel value predMatrix[x of pixel to be predicted y), y]=α EP[(x+1)]+β EP[-(y+1)]+γ EP[0], calculate weight coefficient α, β, γ according to above-mentioned reproducing kernel interpolating function again, and for computer realization, the denominator of weight coefficient is taken as 2 power, so that division arithmetic can be realized with the dextroposition computing.Therefore can obtain under the AVS standard 8 * 8 luminance block and/or 8 * 8 chrominance block and/or 8 * 8 chrominance block under the standard H.264/AVC, and r[i], c[i] but (i=0...9) equal time spent, the calculating formula of DC mode prediction method:

predMatrix[0，0]＝(56×EP[1]+56×EP[-1]-48×EP[0])/64；

predMatrix[0，1]＝(20×EP[1]+29×EP[-2]-17×EP[0])/32；

predMatrix[0，2]＝(24×EP[1]+61×EP[-3]-21×EP[0])/64；

predMatrix[0，3]＝(27×EP[1]+124×EP[-4]-23×EP[0])/128；

predMatrix[0，4]＝(7×EP[1]+63×EP[-5]-6×EP[0])/64；

predMatrix[0，5]＝(6×EP[1]+127×EP[-6]-5×EP[0])/128；

predMatrix[0，6]＝(1×EP[1]×EP[-7]-1×EP[0])/64；

predMatrix[0，7]＝(1×EP[1]×EP[-8]-1×EP[0])/128；

predMatrix[1，0]＝(29×EP[2]+20×EP[-1]-17×EP[0])/32；

predMatrix[1，1]＝(45×EP[2]5×EP[-2]-26×EP[0])/64；

predMatrix[1，2]＝(31×EP[2]+51×EP[-3]-18×EP[0])/64；

predMatrix[1，3]＝(9×EP[2]+28×EP[-4]-5×EP[0])/32；

predMatrix[1，4]＝(77×EP[2]81×EP[-5]-46×EP[0])/512；

predMatrix[1，5]＝(5×EP[2]+62×EP[-6]-3×EP[0])/64；

predMatrix[1，6]＝(10×EP[2]+251×EP[-7]-5×EP[0])/256；

predMatrix[1，7]＝(10×EP[2]+507×EP[-8]-5×EP[0])/512；

predMatrix[2，0]＝(61×EP[3]+24×EP[-1]-21×EP[0])/64；

predMatrix[2，1]＝(51×EP[3]+31×EP[-2]-18×EP[0])/64；

predMatrix[2，2]＝(5×EP[3]+5×EP[-3]-2×EP[0])/8；

predMatrix[2，3]＝(51×EP[3]+95×EP[-4]-18×EP[0])/128；

predMatrix[2，4]＝(15×EP[3]+54×EP[-5]-5×EP[0])/64；

predMatrix[2，5]＝(15×EP[3]+118×EP[-6]-5×EP[0])/128；

predMatrix[2，5]＝(4×EP[3]+61×EP[-7]-1×EP[0])/64；

predMatrix[2，7]＝(4×EP[3]+125×EP[-8]-1×EP[0])/128；

predMatrix[3，0]＝(124×EP[4]+27×EP[-1]-23×EP[0])/128；

predMatrix[3，1]＝(28×EP[4]+9×EP[-2]-5×EP[0])/32；

predMatrix[3，2]＝(95×EP[4]+51×EP[-3]-18×EP[0])/128；

predMatrix[3，3]＝(35×EP[4]+35×EP[-4]-6×EP[0])/64；

predMatrix[3，4]＝(46×EP[4]+91×EP[-5]-9×EP[0])/128；

predMatrix[3，5]＝(27×EP[4]+106×EP[-6]-5×EP[0])/128；

predMatrix[3，6]＝(7×EP[4]+58×EP[-7]-1×EP[0])/64；

predMatrix[3，7]＝(6×EP[4]+123×EP[-8]-1×EP[0])/128；

predMatrix[4，0]＝(63×EP[5]+7×EP[-1]-6×EP[0])/64；

predMatrix[4，1]＝(481×EP[5]+77×EP[-2]-46×EP[0])/512；

predMatrix[4，2]＝(54×EP[5]+15×EP[-3]-5×EP[0])/64；

predMatrix[4，3]＝(91×EP[5]6×EP[-4]-9×EP[0])/128；

predMatrix[4，4]＝(269×EP[5]+269×EP[-5]-2?6×EP[0])/512；

predMatrix[4，5]＝(174×EP[5]+353×EP[-6]-15×EP[0])/512；

predMatrix[4，6]＝(12×EP[5]+53×EP[-7]-1×EP[0])/64；

predMatrix[4，7]＝(13×EP[5]+116×EP[-8]-1×EP[0])/128；

predMatrix[5，0]＝(127×EP[6]+6×EP[-1]-5×EP[0])/128；

predMatrix[5，1]＝(62×EP[6]+5×EP[-2]-3×EP[0])/64；

predMatrix[5，2]＝(118×EP[6]+15×EP[-3]-5×EP[0])/128；

predMatrix[5，3]＝(106×EP[6]+27×EP[-4]-5×EP[0])/128；

predMatrix[5，4]＝(353×EP[6]+174×EP[-5]-15×EP[0])/512；

predMatrix[5，5]＝(33×EP[6]+33×EP[-6]-2×EP[0])/64；

predMatrix[5，6]＝(21×EP[6]4×EP[-7]-1×EP[0])/64；

predMatrix[5，7]＝(23×EP[6]+106×EP[-8]-1×EP[0])/128；

predMatrix[6，0]＝(64×EP[7]+1×EP[-1]-1×EP[0])/64；

predMatrix[6，1]＝(251×EP[7]+10×EP[-2]-5×EP[0])/256；

predMatrix[6，2]＝(61×EP[7]×EP[-3]-1×EP[0])/64；

predMatrix[6，3]＝(58×EP[7]+7×EP[-4]-1×EP[0])/64；

predMatrix[6，4]＝(53×EP[7]+12×EP[-5]-1×EP[0])/6?4；

predMatrix[6，5]＝(44×EP[7]+21×EP[-6]-1×EP[0])/64；

predMatrix[6，6]＝(259×EP[7]+259×EP[-7]-6×EP[0])/512；

predMatrix[6，7]＝(41×EP[7]+88×EP[-8]-1×EP[0])/128；

predMatrix[7，0]＝(128×EP[8]+1×EP[-1]-1×EP[0])/128；

predMatrix[7，1]＝(507×EP[8]+10×EP[-2]-5×EP[0])/512；

predMatrix[7，2]＝(125×EP[8]×EP[-3]-1×EP[0])/128；

predMatrix[7，3]＝(123×EP[8]+6×EP[-4]-1×EP[0])/128；

predMatrix[7，4]＝(116×EP[8]+13×EP[-5]-1×EP[0])/128；

predMatrix[7，5]＝(106×EP[8]+23×EP[-6]-1×EP[0])/128；

predMatrix[7，6]＝(88×EP[8]+41×EP[-7]-1×EP[0])/128；

predMatrix[7，7]＝(1029×EP[8]+1029×EP[-8]-10×EP[0])/2048。

(2) if having only r[i] (i=0...9) available, but promptly has only on the current pixel piece row time spent adjacent in the piece, according to the pixel value of pixel value prediction current pixel piece that is positioned at second pixel of same column in the last piece of current pixel piece with pixel to be predicted with the current pixel piece.The pixel value that is specially this second pixel carries out filtering, obtains second value, with the pixel value of this second value as pixel to be predicted.

If the line number of current pixel piece and columns are all more than or equal to 8 o'clock (present embodiment is this situation), pixel value to this second pixel carries out filtering, obtaining second value is specially: the weight of the pixel value of left pixel that will be adjacent with described second pixel is made as 1, the weight of the pixel value of described second pixel is made as 2, the weight of the pixel value of right pixel that will be adjacent with described second pixel is made as 1, with the mean value of the data after the above-mentioned weighted, as described second value.

Therefore can obtain under the AVS standard 8 * 8 luminance block and/or 8 * 8 chrominance block and/or 8 * 8 chrominance block under the standard H.264/AVC, have only r[i] but (i=0...9) time spent, the calculating formula of DC mode prediction method:

predMatrix[x，y]＝(r[x]+2×r[x+1]+r[x+2]+2)＞＞2(x，y＝0..7)。

(3) if having only c[i] (i=0...9) available, but promptly has only the row time spent adjacent in the left piece of current pixel piece, according to the pixel value that is positioned at the pixel value prediction current pixel piece of first pixel of going together in the left piece of current pixel piece with pixel to be predicted with the current pixel piece.The pixel value that is specially this first pixel carries out filtering, obtains first value, with the pixel value of this first value as pixel to be predicted.

If the line number of current pixel piece and columns are all more than or equal to 8 o'clock (present embodiment is this situation), pixel value to this first pixel carries out filtering, obtaining first value is specially: the weight of the pixel value of last pixel that will be adjacent with described first pixel is made as 1, the weight of the pixel value of described first pixel is made as 2, the weight of the pixel value of following pixel that will be adjacent with described first pixel is made as 1, with the mean value of the data after the above-mentioned weighted, as described first value.

Therefore can obtain under the AVS standard 8 * 8 luminance block and/or 8 * 8 chrominance block and/or 8 * 8 chrominance block under the standard H.264/AVC, have only c[i] but (i=0...9) time spent, the calculating formula of DC mode prediction method:

predMatrix[x，y]＝(c[y]+2×c[y+1]+c[y+2]+2)＞＞2(x，y＝0..7)。

(4), the pixel value of arbitrary pixel in the described current pixel piece all is made as 128 if when the row adjacent with the current pixel piece in the piece is all unavailable on the row adjacent with the current pixel piece in the piece of a current pixel piece left side, current pixel piece.

Therefore can obtain under the AVS standard 8 * 8 luminance block and/or 8 * 8 chrominance block and/or 8 * 8 chrominance block under the standard H.264/AVC, r[i], c[i] when (i=0...9) all unavailable, the calculating formula of DC mode prediction method:

predMatrix[x，y]＝128(x，y＝0..7)。

Second embodiment: before carrying out the DC prediction, will obtain reference sample (pixel) earlier, predict with the foundation reference sample.If the image pattern matrix under the current pixel piece is 4 * 4 matrix I, I represents luminance matrix.If not " existence " or this sample decoding as yet of the piece at certain sample place, this sample " unavailable " then, otherwise this sample " available ".

If the coordinate of current pixel piece upper left corner sample be (x0, y0), wherein, x0 represents row, y0 represents row, then current pixel piece reference sample obtains by following rule:

If coordinate be (promptly the row adjacent with the current pixel piece is available in the last piece of current pixel piece for x0+i-1, the y0-1) sample of (i=1..4) " available ", and then this row element (pixel value of reference sample) is: r[i]=I[y0-1] [x0+i-1]; Otherwise r[i] " unavailable ";

If coordinate be (promptly the row adjacent with the current pixel piece is available in upper right of the current pixel piece, then r[i for x0+i-1, the y0-1) sample of (i=5..8) " available "]=I[y0-1] [x0+i-1]; Otherwise r[i] equal r[4];

If coordinate be (promptly row adjacent with the current pixel piece are available in the left piece of current pixel piece, then c[i for x0-1, the y0+i-1) sample of (i=1..4) " available "]=I[y 0+i-1] [x0-1]; Otherwise c[i] " unavailable ";

If coordinate be (x0-1, sample y0-1) " available ", then r[0]=I[y0-1] [x0-1]; Otherwise:

If r[1] " available ", and c[1] " available ", r[0]=(r[1]+c[1]+1)＞＞1;

If r[1] " available ", and c[1] " unavailable ", r[0]=r[1];

If c[1] " available ", and r[1] " unavailable ", r[0]=c[1];

Otherwise r[0] " unavailable ".

Obtain reference sample, will carry out pixel value DC prediction to the current pixel piece, computational methods are as follows:

(1) if r[i], c[i] (i=0...4) all available, be the row adjacent in the piece of a current pixel piece left side with the current pixel piece, but equal time spent of the row adjacent in the piece on the current pixel piece with the current pixel piece, in row adjacent in the piece of a current pixel piece left side with the current pixel piece, be positioned at the pixel value of first pixel of going together with pixel to be predicted, on the current pixel piece in the piece in the row adjacent with the current pixel piece, be positioned at the pixel value of second pixel of same column with pixel to be predicted, the pixel value of three pixel adjacent in upper left of the current pixel piece with the current pixel piece, the pixel value of the pixel to be predicted in the prediction current pixel piece.The pixel value that is specially described first pixel carries out filtering, obtains first value, and the pixel value of described second pixel is carried out filtering, obtains second value, and the pixel value of described the 3rd pixel is carried out filtering, obtains the 3rd value; Described first value, second value and the 3rd value are weighted processing, with the pixel value of the value after the weighting as pixel to be predicted.

If have one in the line number of current pixel piece and the columns at least at 8 o'clock less than (present embodiment is this situation), pixel value to described first pixel carries out filtering, obtaining first value is specially: the weight of the pixel value of last pixel that will be adjacent with described first pixel is made as 0, the weight of the pixel value of described first pixel is made as 1, the weight of the pixel value of following pixel that will be adjacent with described first pixel is made as 0, with the mean value of the data after the above-mentioned weighted, as described first value; Pixel value to described second pixel carries out filtering, obtaining second value is specially: the weight of the pixel value of left pixel that will be adjacent with described second pixel is made as 0, the weight of the pixel value of described second pixel is made as 1, the weight of the pixel value of right pixel that will be adjacent with described second pixel is made as 0, with the mean value of the data after the above-mentioned weighted, as described second value; Pixel value to described the 3rd pixel carries out filtering, obtaining the 3rd value is specially: the weight of the pixel value of right pixel that will be adjacent with described the 3rd pixel is made as 0, the weight of the pixel value of described the 3rd pixel is made as 1, the weight of the pixel value of following pixel that will be adjacent with described the 3rd pixel is made as 0, with the mean value of the data after the above-mentioned weighted, as described the 3rd value.

Be exemplified below: the current pixel piece with above-mentioned 4 * 4 is an example, as to want the prediction line coordinate be y1, the row coordinate is the pixel value of the pixel of x 1, and the pixel value of first pixel is c[y1+1 so], first value is EP[-(y1+1)]=0 * c[y1]+1 * c[y1+1]+0 * c[y1+2]=c[y1+1]; The pixel value of second pixel is r[x1+1], second value is EP[(x1+1)]=0 * r[x1]+1 * r[x1+1]+0 * r[x1+2]=r[x1+1]; The pixel value of the 3rd pixel is r[0], the 3rd value is EP[0]=0 * r[1]+1 * r[0]+0 * c[1]=r[0]; Above-mentioned first value, second value, the 3rd value are weighted processing again, and the value after the processing is as predicted value, i.e. predicted value predMatrix[x, y]=α 1 EP[(x+1)]+β 1EP[-(y+1)]+γ 1 EP[0]

＝α1r[x+1]+β1?c[-(y+1)]+γ1r[0]

Wherein weight coefficient α 1, β 1, γ 1 obtain according to the reproducing kernel interpolating function, and the formation of reproducing kernel interpolating function is with the principle that forms under the above-mentioned AVS standard.

Therefore can obtain H.2 4 * 4 luminance block under the 6 4/AVC standards, and r[i], c[i] but (i=1...4) equal time spent, the calculating formula of DC mode prediction method:

predMatrix[0，0]＝(7×r[1]+7×c[1]-6×r[0]+4)＞＞3；

predMatrix[1，0]＝(29×r[2]+20×c[1]-17×r[0]+16)＞＞5；

predMatrix[0，1]＝(20×r[1]+29×c[2]-17×r[0]+16)＞＞5；

predMatrix[2，0]＝(15×r[3]+6×c[1]-5×r[0]+8)＞＞4；

predMatrix[0，2]＝(6×r[1]+15×c[3]～5×r[0]+8)＞＞4；

predMatrix[3，0]＝(31×r[4]+7×c[1]-6×r[0]+16)＞＞5；

predMatrix[0，3]＝(7×r[1]+3?1×c[4]-6×r[0]+16)＞＞5；

predMatrix[2，1]＝(103×r[3]+61×c[2]-36×r[0]+64)＞＞7；

predMatrix[1，2]＝(61×r[2]+103×c[3]-36×r[0]+64)＞＞7；

predMatrix[3，1]＝(28×r[4]+9×c[2]-5×r[0]+16)＞＞5；

predMatrix[1，3]＝(9×r[2]+28×c[4]-5×r[0]+16)＞＞5；

predMatrix[3，2]＝(47×r[4]+26×c[3]-9×r[0]+32)＞＞6；

predMatrix[2，3]＝(26×r[3]+47×c[4]-9×r[0]+32)＞＞6；

predMatrix[1，1]＝(45×r[2]+45×c[2]-26×r[0]+32)＞＞6；

predMatrix[2，2]＝(77×r[3]+77×c[3]-26×r[0]+64)＞＞7；

predMatrix[3，3]＝(35×r[4]+35×c[4]-6×r[0]+32)＞＞6。

(2) if having only r[i] (i=1...4) available, but promptly has only on the current pixel piece row time spent adjacent in the piece, according to the pixel value of pixel value prediction current pixel piece that is positioned at second pixel of same column in the last piece of current pixel piece with pixel to be predicted with the current pixel piece.The pixel value that is specially this second pixel carries out filtering, obtains second value, with the pixel value of this second value as pixel to be predicted.

If the line number of current pixel piece and columns have one at least less than (present embodiment is this situation) at 8 o'clock, pixel value to this second pixel carries out filtering, obtaining second value is specially: the weight of the pixel value of left pixel that will be adjacent with described second pixel is made as 0, the weight of the pixel value of described second pixel is made as 1, the weight of the pixel value of right pixel that will be adjacent with described second pixel is made as 0, with the mean value of the data after the above-mentioned weighted, as described second value.

Therefore can obtain H.264/AVC 4 * 4 luminance block under the standard, have only r[i] but (i=1...4) time spent, the calculating formula of DC mode prediction method:

predMatrix[x，y]＝(r[1]+r[2]+r[3]+r[4]+2)＞＞2。

(3) if having only c[i] (i=1...4) available, but promptly has only the row time spent adjacent in the left piece of current pixel piece, according to the pixel value that is positioned at the pixel value prediction current pixel piece of first pixel of going together in the left piece of current pixel piece with pixel to be predicted with the current pixel piece.The pixel value that is specially this first pixel carries out filtering, obtains first value, with the pixel value of this first value as pixel to be predicted.

If the line number of current pixel piece and columns have one at least less than (present embodiment is this situation) at 8 o'clock, pixel value to this first pixel carries out filtering, obtaining first value is specially: the weight of the pixel value of last pixel that will be adjacent with described first pixel is made as 0, the weight of the pixel value of described first pixel is made as 1, the weight of the pixel value of following pixel that will be adjacent with described first pixel is made as 0, with the mean value of the data after the above-mentioned weighted, as described first value.

Therefore can obtain H.264/AVC 4 * 4 luminance block under the standard, have only c[i] but (i=1...4) time spent, the calculating formula of DC mode prediction method:

predMatrix[x，y]＝(c[1]+c[2]+c[3]+c[4]+2)＞＞2。

(4), the pixel value of arbitrary pixel in the described current pixel piece all is made as 128 if when the row adjacent with the current pixel piece in the piece is all unavailable on the row adjacent with the current pixel piece in the piece of a current pixel piece left side, current pixel piece.

Therefore can obtain H.264/AVC 4 * 4 luminance block under the standard, r[i], c[i] when (i=1...4) all unavailable, the calculating formula of DC mode prediction method:

predMatrix[x，y]＝128(x，y＝0..3)。

It should be noted that at last: above embodiment only in order to technical scheme of the present invention to be described, is not intended to limit; Although with reference to previous embodiment the present invention is had been described in detail, those of ordinary skill in the art is to be understood that: it still can be made amendment to the technical scheme that aforementioned each embodiment put down in writing, and perhaps part technical characterictic wherein is equal to replacement; And these modifications or replacement do not make the essence of appropriate technical solution break away from the spirit and scope of various embodiments of the present invention technical scheme.

Claims (9)

1. DC mode prediction technique is characterized in that comprising:

Choose the macro block of current prediction, this macro block is divided into several block of pixels;

Choose a block of pixels as the current pixel piece, in the left pixel piece of this current pixel piece with this current pixel piece adjacent pixels row in, the pixel that is positioned at delegation with pixel to be predicted in this current pixel piece is first pixel, in the upside block of pixels of this current pixel piece with this current pixel piece adjacent pixels row in, the pixel that is positioned at same row with pixel to be predicted in this current pixel piece is second pixel, in the upper left side block of pixels of this current pixel piece, with this current pixel piece adjacent pixels point be the 3rd pixel; If the pixel value of all pixels in the pixel column at the pixel column at the described first pixel place and the second pixel place, and the pixel value of the 3rd pixel is all known, then described first pixel is carried out Filtering Processing and obtain first value, described second pixel is carried out Filtering Processing obtain second value, described the 3rd pixel is carried out Filtering Processing obtain the 3rd value; Described first value, second value and the 3rd value be weighted handle the pixel value obtain described pixel to be predicted;

If have only the pixel value of the pixel in the pixel column at the described first pixel place known, then will carry out first value that obtains after the Filtering Processing pixel value to described first pixel as described pixel to be predicted;

If have only the pixel value of the pixel in the pixel column at the described second pixel place known, then will carry out second value that obtains after the Filtering Processing pixel value to described second pixel as described pixel to be predicted;

When if the pixel value of all pixels in the pixel column at the pixel column at the described first pixel place and the second pixel place is all unknown, the pixel value of arbitrary pixel in the described current pixel piece is made as 128.

2. DC mode prediction technique according to claim 1 is characterized in that, described macro block is 16 * 16 pixel macro blocks.

3. DC mode prediction technique according to claim 1 is characterized in that, describedly this macro block is divided into several block of pixels is specially: with this macro block by from left to right, from top to bottom order is divided into several block of pixels of ordered series of numbers and number row.

4. DC mode prediction technique according to claim 1 is characterized in that the line number of described current pixel piece and columns are all more than or equal to 8.

5. DC mode prediction technique according to claim 4 is characterized in that,

Describedly described first pixel is carried out Filtering Processing obtain first value and be specially: on the pixel column with the described first pixel place, the weight that is positioned at the pixel value of the described first pixel upside adjacent pixels point is made as 1, the weight of the pixel value of described first pixel is made as 2, on the pixel column with the described first pixel place, the weight that is positioned at the pixel value of the described first pixel downside adjacent pixels point is made as 1, with the mean value of the data after the above-mentioned weighted, as described first value;

Describedly described second pixel is carried out Filtering Processing obtain second value and be specially: on the pixel column with the described second pixel place, the weight that is positioned at the pixel value of described second pixel left side adjacent pixels point is made as 1, the weight of the pixel value of described second pixel is made as 2, on the pixel column with the described second pixel place, the weight that is positioned at the pixel value of the described second pixel right side adjacent pixels point is made as 1, with the mean value of the data after the above-mentioned weighted, as described second value;

Describedly described the 3rd pixel is carried out Filtering Processing obtain the 3rd value and be specially: on the pixel column with described the 3rd pixel place, the weight that is positioned at the pixel value of described the 3rd pixel right side adjacent pixels point is made as 1, the weight of the pixel value of described the 3rd pixel is made as 2, on the pixel column with described the 3rd pixel place, the weight that is positioned at the pixel value of described the 3rd pixel downside adjacent pixels point is made as 1, with the mean value of the data after the above-mentioned weighted, as described the 3rd value.

6. DC mode prediction technique according to claim 1 is characterized in that, has one in the line number of described current pixel piece and the columns at least less than 8.

7. DC mode prediction technique according to claim 6 is characterized in that,

Describedly described first pixel is carried out Filtering Processing obtain first value and be specially: on the pixel column with the described first pixel place, the weight that is positioned at the pixel value of the described first pixel upside adjacent pixels point is made as 0, the weight of the pixel value of described first pixel is made as 1, on the pixel column with the described first pixel place, the weight that is positioned at the pixel value of the described first pixel downside adjacent pixels point is made as 0, with the mean value of the data after the above-mentioned weighted, as described first value;

Describedly described second pixel is carried out Filtering Processing obtain second value and be specially: on the pixel column with the described second pixel place, the weight that is positioned at the pixel value of described second pixel left side adjacent pixels point is made as 0, the weight of the pixel value of described second pixel is made as 1, on the pixel column with the described second pixel place, the weight that is positioned at the pixel value of the described second pixel right side adjacent pixels point is made as 0, with the mean value of the data after the above-mentioned weighted, as described second value;

Describedly described the 3rd pixel is carried out Filtering Processing obtain the 3rd value and be specially: on the pixel column with described the 3rd pixel place, the weight that is positioned at the pixel value of described the 3rd pixel right side adjacent pixels point is made as 0, the weight of the pixel value of described the 3rd pixel is made as 1, on the pixel column with described the 3rd pixel place, the weight that is positioned at the pixel value of described the 3rd pixel downside adjacent pixels point is made as 0, with the mean value of the data after the above-mentioned weighted, as described the 3rd value.

8. DC mode prediction technique according to claim 1 is characterized in that, described described first value, second value and the 3rd value is weighted the weight coefficient of processing, obtains by the reproducing kernel interpolating function.

9. DC mode prediction technique according to claim 8 is characterized in that, the interpolation knot of described reproducing kernel interpolating function is and the known pixel of current pixel piece adjacent pixels value.

Images