CN1816149A - Filtering-wave method of removing blocking-effect in video-frequency picture and loop filter - Google Patents

Abstract

The method includes two parts: filtering decision and filtering execution. The filtering decision includes steps: based on pixels of received image and coding information to determine whether there is real edge existed on borderline of two adjacent small blocks; determining value of block effect of each two adjacent small blocks; there is one-to-one correspondence between different values of block effect and different filtering strengths. The filtering execution includes steps: not carrying out filtering operation for adjacent small blocks of containing real edge on borderline; carrying out filtering operation for adjacent small blocks of not containing real edge on borderline by using selected filtering method corresponding to filtering strength based on values of block effect. Filtering inputted video images, the disclosed method removes block effect caused by blocked encoding so as to raise quality of video image.

Description

Remove the filtering method and the loop filter of blocking effect in the video image

Technical field

The present invention relates to the video image smothing filtering technology in multimedia field, specially refer to a kind of loop filter of removing the filtering method of blocking effect in the video image and realizing this method.

Background technology

In the video compression coding process, in order to eliminate the spatial coherence of vedio data frame, usually the pixel of each frame is divided into the block of pixels of several N * N, be called macro block, generally speaking, N is 16.Wherein, each macro block can also further be divided into 1 16 * 16 or several fritters of 8 * 8,16 * 8,8 * 16 or 4 * 4, and in the encoding process process of video image, is the base unit of video coding and estimation with the fritter of being divided.Above-mentionedly thisly pixel is carried out the method that block encoding handles to reduce vedio data frame information redundancy spatially, realize the compression of transmitting video image desired data amount.But, because the quantization operation of introducing in video compression is a lossy compression method, thereby causes the blocking effect of video image easily, promptly in the image that receives usually said " mosaic " phenomenon may appear, cause the subjective quality of video image to descend.Therefore, in the process of image processing, must there be a kind of method of removal of images blocking effect, to improve the quality of the video image that the user receives.

Summary of the invention

In order to solve the problems of the technologies described above, the invention provides a kind of method of removing blocking effect in the video coding, can eliminate video image because block encoding is handled the blocking effect that is caused, improve the quality of user institute receiver, video image.

The present invention also provides a kind of loop filter, can use the method for removal video image blocking effect of the present invention, removes the blocking effect in the inputted video image, realizes the real-time coding and decoding of video image.

The filtering method of blocking effect in the removal video image of the present invention comprises: two steps are carried out in filtering judgement and filtering, wherein,

Described filtering judgement is: according to the pixel and the coded message that receive image, determine whether to exist on the border of per two adjacent isles the true edge of image and the blocking effect value of determining per two adjacent isles, wherein, different blocking effect values is corresponding one by one with different filtering dynamics;

Described filtering is implemented as: for the adjacent isles that has true edge on the border, do not carry out filtering; For the adjacent isles that on the border, does not have true edge,, select the filtering method corresponding to carry out filtering with this filtering dynamics according to the filtering dynamics of its blocking effect value correspondence.

Whether exist the true edge of image to comprise on the border of definite two adjacent isles of the present invention:

A, according to the pixel that receives image, calculate pixel value difference absolute value in the piece of each fritter in the interblock pixel value difference absolute value of adjacent isles and this adjacent isles respectively;

B, pixel value difference thresholding in pixel value difference absolute value and predefined interblock pixel value difference thresholding and the piece in the interblock pixel value difference absolute value that calculates and the piece is compared, if described interblock pixel value difference absolute value is greater than described interblock pixel value difference thresholding, then there is the true edge of image in the interior pixel value difference absolute value of piece that a fritter is arranged in the perhaps described adjacent isles greater than described interior pixel value difference thresholding on the border of two fritters; Otherwise, do not have the true edge of image on the border of two adjacent isles.

The blocking effect value of definite two adjacent isles of the present invention is: the coded message that receives image judged,

When any fritter in described two adjacent isles is the macroblock edges fritter of intraframe coding, determine the blocking effect value of blocking effect value for corresponding strong filtering;

When the macroblock edges fritter that described two adjacent isles all are not intraframe codings, when still having the fritter of motion prediction difference or code coefficient difference, determine the blocking effect value of blocking effect value for corresponding weak filtering;

When above condition does not satisfy, determine the blocking effect value of blocking effect value for corresponding not filtering.

The macroblock edges fritter of intraframe coding of the present invention is to belong to the infra-frame prediction macro block and its border is the fritter of affiliated infra-frame prediction macroblock boundaries;

Described two adjacent isles all are not the macroblock edges fritters of intraframe coding, but having the fritter of motion prediction difference or code coefficient difference is the fritter that satisfies one of following four conditions:

Belong to the infra-frame prediction macro block, but its border not the border of affiliated macro block;

Code coefficient is not zero entirely;

The reference frame that is adjacent fritter is inequality;

The reference frame that is adjacent fritter is identical, and still the difference with the adjacent isles motion vector has one-component at least greater than a pixel.

The blocking effect value of the weak filtering of correspondence of the present invention further is included as 3 sub-grades;

Definite method of the sub-grade of blocking effect value of the weak filtering of described correspondence is as follows:

Any fritter in described two adjacent isles belongs to the infra-frame prediction macro block, but the little block boundary that belongs to the infra-frame prediction macro block determines that the sub-grade of blocking effect value of corresponding weak filtering is the first sub-grade of weak filtering when not being the border of macro block under it;

When the code coefficient that has a fritter in described two adjacent isles at least is not zero entirely, determine that the sub-grade of blocking effect value of corresponding weak filtering is the second sub-grade;

Reference frames inequality when the reference frame of described two adjacent isles or described two adjacent isles are identical, when but the difference of described two little block motion vectors has one-component greater than a pixel at least, determine that the sub-grade of blocking effect value of corresponding weak filtering is the 3rd sub-grade.

Filtering dynamics according to its blocking effect value correspondence of the present invention, select the filtering method corresponding to be filtered into this filtering dynamics:

When the blocking effect value is the blocking effect value of corresponding not filtering, do not carry out filtering, directly the image and the coded message of output reception;

When the blocking effect value is the blocking effect value of corresponding weak filtering, the image that receives is all carried out the difference filtering corresponding with weak filtering dynamics;

When the blocking effect value is the blocking effect value of corresponding strong filtering, the image that receives is carried out and the strong corresponding mean filter of filtering dynamics.

The method of difference filtering of the present invention is as follows:

A1, calculate according to the pixel on both sides, adjacent isles border each pixel in the filter window is adjusted required a small amount of;

B1, according to the arithmetic mean value of the quantization parameter of adjacent two fritters, in the pixel increase and decrease difference threshold table of setting up in advance relevant, search corresponding pixel increase and decrease difference threshold, and a small amount of that step a1 calculates is limited in searches the pixel that obtains and increase and decrease in the difference threshold scope with quantization parameter;

C1, with each pixel in the filter window and its corresponding addition in a small amount through step b1 restriction, obtain the pixel behind interior each pixel filter of filter window.

If P1, P0 and Q0, Q1 are tactic pixel on two adjacent isles borders, wherein, P0 and Q0 are pixel in the described filter window, and a small amount of of its correspondence is Delta-P0, Delta-Q0;

Described being calculated as of step a uses weight coefficient (1 ,-2,2 ,-1) to calculate Delta-P0 and Delta-Q0 in a small amount.

If P2, Q2 be on two adjacent isles borders with P1 or Q1 adjacent pixels, P1 and Q1 also are pixel in the described filter window, a small amount of of its correspondence is Delta-P1, Delta-Q1;

Described being calculated as of step a uses weight coefficient (2 ,-3,1) to calculate Delta-P1 and Delta-Q1 in a small amount.

Difference filtering method of the present invention is as follows:

A2, calculate according to the pixel on both sides, adjacent isles border each pixel in the filter window is adjusted required a small amount of;

B2, according to the sub-grade of blocking effect value of the corresponding weak filtering of adjacent two fritters, in pixel increase and decrease difference threshold table that set up in advance and the sub-rank correlation of blocking effect value, searching corresponding pixel increase and decrease difference threshold, and a small amount of that step a2 calculates is limited in searching in the pixel increase and decrease difference threshold scope that obtains;

C2, with each pixel in the filter window and its corresponding addition in a small amount through step b2 restriction, obtain the pixel behind interior each pixel filter of filter window.

If P2, P1, P0 and Q0, Q1, Q2 are tactic pixel on two adjacent isles borders, wherein, P1, P0 and Q0, Q1 are pixel in the described filter window, and a small amount of of its correspondence is Delta-P0, Delta-P1, Delta-Q0 and Delta-Q1;

Step a1 is described to be calculated as: use weight coefficient (1 ,-4,4 ,-1) to calculate Delta-P0, Delta-Q0 in a small amount, use weight coefficient (2 ,-4,1,1) to calculate Delta-P1, Delta-Q1 in a small amount.

Mean filter of the present invention is: the pixel according to both sides, adjacent isles border recomputates each point pixel in the filter window.

If P1, P0 and Q0, Q1 are tactic pixel on two adjacent isles borders, wherein, P0 and Q0 are pixel in the described filter window;

Described being calculated as: use weight coefficient (1,2,1) calculating pixel P0 and Q0.

If P1 and Q1 also are pixel in the described filter window;

Described being calculated as: use weight coefficient (1,2,1) calculating pixel P1 and Q1.

If P3, P2, P1, P0 and Q0, Q1, Q2, P3 are tactic pixel on two adjacent isles borders, wherein, P2, P1, P0 and Q0, Q1, Q2 are pixel in the described filter window;

Described being calculated as: use weight coefficient (1,2,2,2,1) calculating pixel P0, Q0, use weight coefficient (1,1,1,1) calculating pixel P1, Q1, use weight coefficient (2,3,1,1,1) calculating pixel P2, Q2.

The loop filter of blocking effect in the removal video image of the present invention comprises:

Filtering decision unit (100) is used for the image of input is carried out rim detection, and the blocking effect value of determining two adjacent isles, and wherein, different blocking effect values is corresponding one by one with different filtering dynamics;

Filtering performance element (200) is used for judging according to the edge detection results of filtering decision unit (100) input and definite blocking effect value, for the adjacent isles that has true edge on the border, does not carry out filtering; For the adjacent isles that on the border, does not have true edge,, select the filtering method corresponding to carry out filtering with this filtering dynamics according to the filtering dynamics of its blocking effect value correspondence.

Filtering decision unit of the present invention (100) further comprises:

Pixel and coded message memory module (110) are used to receive and the pixel of the pending image that buffer memory is imported and the coded message of adjacent two fritters, and output;

Threshold value determination module (120), be used for according to quantization parameter from pixel and two fritters of coded message memory module (110) coded message, be identified for carrying out the interblock pixel value difference thresholding and the interior pixel value difference thresholding of piece of rim detection, and these two threshold values are outputed to rim detection module (150);

Pixel value difference computing module (130) is used for according to the pixel value difference absolute value that calculates pixel on the adjacent isles border from the pixel of pixel and coded message memory module (110), and the absolute difference that calculates is outputed to rim detection module (150);

Rim detection module (150) is used for threshold value and pixel value difference absolute value according to input, adjudicates the true edge that whether has image on the described adjacent isles border, and court verdict is outputed to filtering performance element (200);

Blocking effect value computing module (140), be used for according to from the pixel of pixel and coded message memory module (110) and the blocking effect value that coded message is calculated described adjacent isles, and the blocking effect value determined and described pixel and coded message are outputed to filtering performance element (200).

Filtering performance element of the present invention (200) further comprises:

Filtering trigger module (210), be used for carrying out the filtering judgement according to edge detection results and blocking effect value from described filtering decision unit (100), to the filtering dynamics that has image true edge or blocking effect value correspondence on the adjacent isles border is the fritter of not filtering, and the pixel and the coded message of input are directly exported; Filtering dynamics to blocking effect value correspondence is the fritter of weak filtering, and pixel and the coded message of importing outputed to difference filtration module (220); Filtering dynamics to blocking effect value correspondence is the fritter of strong filtering, with the pixel and the coded message output mean filter module (230) of input;

Difference filtration module (220) is used for the pixel and the coded message of input are carried out difference filtering;

Mean filter module (230) is used for the pixel and the coded message of input are carried out mean filter.

This shows, generally speaking, use the method for blocking effect in the removal video coding of the present invention and loop filter and can remove blocking effect in the video image, improve the video image quality that the user receives, realize the real-time coding of video image.

Description of drawings

Fig. 1 is the pixel schematic diagram on both sides, two adjacent isles borders in the video image;

Fig. 2 has shown the structure of loop filter of the present invention.

Embodiment

For the purpose, technical scheme and the advantage that make invention is clearer, below with reference to the accompanying drawing embodiment that develops simultaneously, the present invention is described in further detail.

In order to remove the blocking effect in the video image, the invention provides a kind of method of removing blocking effect in the video image, this method adopts the The disposal of gentle filter method that the pixel on both sides, division fritter border is handled, to remove the blocking effect between the fritter.Fig. 1 is the pixel schematic diagram on both sides, two adjacent isles borders in the video image.Each lattice among Fig. 1 is represented a pixel in the video image, and as shown in Figure 1, described video image has been divided into two fritters at P0 and Q0 point.The pixel on these both sides, two fritter borders is called after P3, P2, P1, P0 and Q0, Q1, Q2, Q3 successively.Described smooth filtering method is exactly that the pixel on both sides, above-mentioned border is carried out The disposal of gentle filter, to reach the purpose of removing the video image blocking effect.

The filtering method of blocking effect mainly comprises in the removal video image of the present invention: two steps are carried out in filtering judgement and filtering.

Wherein, described filtering judgement is specially: according to the pixel and the coded message that receive image, determine whether to exist on the border of per two adjacent isles the true edge of image, determine the blocking effect value of per two adjacent isles simultaneously, wherein, the different blocking effect value of adjacent isles is one to one with different filtering dynamics.

Described filtering is carried out and is specially: for the adjacent isles that has true edge on the border, do not carry out filtering; For the adjacent isles that on the border, does not have true edge,, select the filtering method corresponding to carry out filtering with this filtering dynamics according to the filtering dynamics of its blocking effect value correspondence.

To be described in further detail by the method for the preferred embodiments of the present invention below removal video image blocking effect of the present invention.

Embodiment 1:

The described method of present embodiment is divided into 5 grades with the blocking effect between the adjacent isles, and use blocking effect value Bs parameter is represented the blocking effect grade between the adjacent isles, this parameter range is 0 to 4, wherein, the Bs value is big more to show that the blocking effect between the adjacent isles is serious more, need carry out strong more The disposal of gentle filter.

The described filtering decision method of present embodiment may further comprise the steps:

Step 11: the pixel according to two adjacent isles importing is carried out rim detection to two adjacent isles of input, whether judgement exists real edge on the border of described adjacent isles, if there is true edge, then judge the adjacent isles execution filtering that does not need input, finish described filtering judgement flow process then; Otherwise, execution in step 12.

In this step, described rim detection may further comprise the steps:

The absolute value of a, calculating both sides, described adjacent isles border pixel value difference;

The pixel value difference one that is calculated has three, comprises the difference of a pair of pixel that adjacent isles border narrowing toward each end interblock is adjacent, i.e. interblock pixel value difference, and as pixel P0 and Q0 difference on two little block boundaries among Fig. 1, the absolute value of this difference is C1; The difference that also comprises each a pair of pixel that two fritter inner distance border, both sides, adjacent isles border is nearest, i.e. pixel value difference in the piece, as P0 and P1 difference, its absolute value is C2, and Q0 and Q1 difference, its absolute value is C3;

B, pixel value difference thresholding in the difference that calculates and predefined interblock pixel value difference thresholding and the piece is compared, whether the difference of judging pixel between a described fritter is greater than described interblock pixel value difference thresholding, whether the difference of pixel has one greater than described interior pixel value difference thresholding in perhaps described two fritters, if, think that then there is the true edge of image in these two fritter boundaries, not blocking effect, thereby can not carry out filtering, to avoid smoothly falling the details of image; Otherwise, judge that there is not the true edge of image in two little block boundaries, may need to carry out filtering.

In this step, the pixel value difference thresholding is by two tables in described predefined interblock pixel value difference threshold value and the piece, and promptly α table and β table are determined.Wherein, the α table record under different quantization step QP, the interblock difference threshold value of adjacent isles border narrowing toward each end interblock neighbor, the β table record under different quantization step QP, both sides, adjacent isles border fritter interior pixels difference threshold value.Before comparing judgement, the mean value that needs at first be got the QP value of described two adjacent isles behind the arithmetic mean removes the described α of index as benchmark, the β table, threshold value corresponding with current quantization step QP in the α that retrieves table and the β table as difference threshold β in interblock difference threshold α that differentiates and the piece, and then is carried out rim detection.

The method of described rim detection is specially: the interblock pixel value difference thresholding α corresponding with current quantization step QP in the absolute value of pixel value difference between the fritter that calculates is shown less than α, and less than in the piece corresponding in the β table during pixel value difference thresholding β, then the border of these two fritters does not comprise the true edge of image to the absolute value of two fritter interior pixels differences yet; Otherwise the border of these two fritters then comprises the true edge of image.

For example, the P0 among Fig. 1 and Q0 are subtracted each other obtain in absolute difference C1 and the α table comparing with the corresponding threshold value α of corresponding quantization step QP; P0 and P1 are subtracted each other the absolute value C2 that obtains difference, and Q0 and Q1 subtracted each other the absolute value C3 that obtains difference, with compare with the corresponding threshold value of corresponding quantization step QP in the β table, if C1 is less than α, C2 is less than β, and C3 is less than β, and then there is not the true edge of image in the border of these two fritters, need carry out filtering; Otherwise, then there is the true edge of image, do not need to carry out filtering.

Step 12: for the adjacent isles that may need to carry out Filtering Processing, coded message according to two adjacent isles importing, comprise: macroblock encoding type under the fritter, the motion vector of fritter, reference frame, the blocking effect value Bs of current adjacent two fritters of quantization parameter QP and code coefficient or the like calculation of parameter determines the filtering grade of described adjacent isles.

The computational methods of described blocking effect value Bs are as follows:

Macroblock encoding type under described adjacent two fritters is judged, if have at least one to be the fritter of intraframe coding in described adjacent two fritters, promptly belonging to infra-frame prediction macro block and its border also is the border of affiliated macro block, and then blocking effect value Bs equals 4;

If there is one to belong to infra-frame prediction macro block or two and all belong to the infra-frame prediction macro block in described two adjacent isles, but the little block boundary that belongs to the infra-frame prediction macro block is not the border of macro block under it, and then blocking effect value Bs equals 3;

If having the code coefficient of a fritter in described two adjacent isles at least is not zero entirely, then blocking effect value Bs equals 2;

If the reference frame of described two adjacent isles reference frames inequality or described two adjacent isles are identical, but the difference of described two little block motion vectors has one-component at least greater than a pixel, and then blocking effect value Bs equals 1;

If above all conditions do not satisfy, then blocking effect value Bs equals 0.

After having obtained the blocking effect value Bs of two adjacent isles, just can select corresponding filtering method that the input fritter that does not have the image true edge is finished the filtering execution in step according to blocking effect value Bs.Described filtering is carried out and is specially: for there not being the image true edge, the fritter that may need Filtering Processing, determine the implementation dynamics of filtering according to its blocking effect value Bs: when blocking effect value Bs is 0, do not carry out filtering, directly the image and the coded message of output reception; When blocking effect value Bs is 1 to 3, the image that receives is carried out weak filtering; When the blocking effect value is 4, the image that receives is carried out strong filtering.

Below with reference to Fig. 1, be example with the horizontal direction, describe described weak filtering of present embodiment and strong filtering method in detail.

Being filtered into the filter window width a little less than described is 4, and tap number is 4 difference filtering, and described difference filtering will be recomputated the value that obtains adjacent isles boundary pixel P1, P0, Q0 and Q1, and computational methods are as follows:

At first, pixel P2, P1, P0, Q0, Q1 and Q2 according to both sides, adjacent isles border calculate a small amount of Delta-P1, Delta-P0, Delta-Q0 and the Delta-Q1 that adjusts described pixel P1, P0, Q0 and Q1 respectively, wherein, described being calculated as used weight coefficient (1 ,-4,4,-1) calculates Delta-P0, Delta-Q0 in a small amount, use weight coefficient (2 ,-4,1,1) calculates Delta-P1, Delta-Q1 in a small amount.Specifically, the computing formula of described a small amount of Delta-P1, Delta-P0, Delta-Q0 and Delta-Q1 is as follows:

$\mathrm{Delta}-P1=\frac{P2\×2+P0+Q0-P1\×4}{4};$

$\mathrm{Delta}-P0=\frac{(Q0-P0)\×4+P1-Q1}{8};$

$\mathrm{Delta}-Q0=\frac{(P0-Q0)\×4+Q1-P1}{8};$

$\mathrm{Delta}-Q1=\frac{Q2\×2+Q0+P0-Q1\×4}{4}.$

Then, again according to the blocking effect value Bs of adjacent two fritters, this moment, the span of Bs was 1 to 3, in the relevant pixel increase and decrease difference threshold table Clip1 of the blocking effect value of setting up in advance, search corresponding pixel increase and decrease difference threshold clip1 with adjacent isles, it also is the thresholding of described a small amount of, search in the pixel increase and decrease difference threshold scope [clip1, clip1] that obtains being limited in a small amount.

At last, again the value of pixel P1, P0, Q0 and Q1 is sued for peace respectively with a small amount of Delta-P1, the Delta-P0, Delta-Q0 and the Delta-Q1 that calculate and limit through pixel increase and decrease difference threshold clip1, obtain the pixel of P1, P0 after the final filtering, Q0 and Q1.

The described filter window width that is filtered into by force is 6, and tap number is 5 mean filter, in filtering, use weight coefficient (1,2,2,2,1) recomputate pixel P0, Q0, use weight coefficient (1,1,1,1) recomputates pixel P1, Q1, use weight coefficient (2,3,1,1,1) recomputates pixel P2, Q2.Specifically, described mean filter will adopt following formula to recomputate the value of adjacent isles boundary pixel P1, P0, Q0 and Q1:

$P2=\frac{P3\×2+P2\×3+P1+P0+Q0}{8};$

$P1=\frac{Q0+P0+P1+P2}{4};$

$P0=\frac{Q1+Q0\×2+P0\×2+P1\×2+P2}{8};$

$Q0=\frac{P1+P0\×2+Q0\×2+Q1\×2+Q2}{8};$

$Q1=\frac{P0+Q0+Q1+Q2}{4};$

$Q2=\frac{Q3\×2+Q2\×3+Q1+Q0+P0}{8}.$

By the aforementioned calculation formula, just can obtain pixel value, thereby remove the blocking effect in the video image through both sides, adjacent isles border behind difference or the mean filter.It will be appreciated by those skilled in the art that; the aforementioned calculation formula only is the preferable way of realization of present embodiment; except above-mentioned formula, can also adopt any other suitable computing formula to calculate the pixel on both sides, fritter border, and can not exceed the scope that this invention is intended to protect.

Need to prove that because the fritter of video image comprises horizontal boundary and vertical boundary, therefore, the filtering of video image also is divided into horizontal filtering and vertical filtering both direction.Though above be the filtering method that example has illustrated adjacent isles with the horizontal direction only, the smooth filtering method of vertical direction is also basic identical.

This shows, by above-mentioned filtering judgement and filtering execution in step, can select the filtering method of different dynamics that the image of importing is carried out The disposal of gentle filter according to blocking effect value between adjacent two fritters, reach the purpose of blocking effect in effective removal video image.But,, make that the complexity of filtering algorithm is higher, thereby cause the amount of calculation of filtering very big because the filter window width of above-mentioned filtering method is big, tap is more.Have statistics to show that the amount of calculation of decoder end smothing filtering has accounted for 20% of entire decoder amount of calculation greatly, big like this amount of calculation will expend the computing capability of too many CPU or DSP, implement the comparison difficulty.

Filtering execution in step from the foregoing description 1, in the described filtering method of present embodiment, to the blocking effect value is that 1 to 3 adjacent isles all adopts the difference filtering method of corresponding weak filtering dynamics to handle, only be 4 adjacent isles, just adopt the mean filter method of corresponding strong filtering dynamics to handle the blocking effect value.And in carrying out the difference filtering, the blocking effect value of adjacent isles only is used to limit a small amount of that pixel is adjusted, and the dynamics of filtering algorithm when not determining that filtering is carried out.In view of the above, can simplify above-mentioned filtering algorithm by the method that merges the blocking effect value of filtering dynamics a little less than the above-mentioned correspondence, to reduce the amount of calculation of above-mentioned filtering.

Embodiment 2:

In order to address the above problem, reduce the complexity and the amount of calculation of smothing filtering algorithm, present embodiment provides a kind of filtering method of removal video image blocking effect of simplification, and embodiment 1 described filtering method is simplified.This method is divided into Three Estate with the blocking effect between the adjacent isles, uses blocking effect value Bs parameter to represent equally, and this parameter range from 0 to 2 corresponds respectively to smothing filtering dynamics, i.e. not filtering, weak filtering and the strong filtering of Three Estate.Wherein, to the adjacent isles that blocking effect value Bs is 0, will not carry out filtering; To blocking effect value Bs is 1 adjacent isles, will carry out weak filtering, and corresponding to the adjacent isles that blocking effect value Bs is 2, will carry out strong filtering.

Filtering method of the present invention also mainly comprises two steps of filtering judgement and filtering execution.To at first describe filtering decision method of the present invention in detail below.

Filtering decision method of the present invention may further comprise the steps:

Step 21: the pixel according to two adjacent isles importing is carried out rim detection to two adjacent isles of input, whether judgement exists real edge on the border of described adjacent isles, if there is true edge, then judge the adjacent isles execution filtering that does not need input, finish described filtering judgement flow process then; Otherwise, execution in step 22.

In this step, described rim detection is identical with embodiment 1 described edge detection method.

Step 22: for the adjacent isles that may need to carry out Filtering Processing, coded message according to two adjacent isles importing, comprise: macroblock encoding type under the fritter, the motion vector of fritter, reference frame, the blocking effect value Bs of current adjacent two fritters of quantization parameter QP and code coefficient or the like calculation of parameter determines the filtering grade of described adjacent isles.

The computational methods of described blocking effect value Bs are as follows:

Macroblock encoding type under described adjacent two fritters is judged, if have at least one to be the fritter of intraframe coding in described adjacent two fritters, promptly belong to the infra-frame prediction macro block, and its border also is the border of affiliated macro block, then blocking effect value Bs equals 2;

Otherwise, if the macroblock edges fritter that described two adjacent isles all are not intraframe codings, but there is the fritter of motion prediction difference or code coefficient difference, promptly satisfy any in following four conditions, then blocking effect value Bs equals 1:

Have one to belong to infra-frame prediction macro block or two and all belong to the infra-frame prediction macro block in A, described two adjacent isles, but the little block boundary that belongs to the infra-frame prediction macro block is not the border of macro block under it;

The code coefficient that has a fritter in B, described two adjacent isles at least is not zero entirely;

The reference frame of C, described two adjacent isles is inequality;

The reference frame of D, described two adjacent isles is identical, but the difference of described two little block motion vectors has one-component at least greater than a pixel;

If above all conditions do not satisfy, then blocking effect value Bs equals 0.

After having obtained the blocking effect value Bs of two adjacent isles, just can select corresponding filtering method that the input fritter that does not have the image true edge is finished the filtering execution in step according to blocking effect value Bs.Described filtering is carried out and is specially: for there not being the image true edge, the fritter that may need Filtering Processing is determined the implementation dynamics of filtering according to its blocking effect value Bs, when blocking effect value Bs is 0, do not carry out filtering, directly the image and the coded message of output reception; When blocking effect value Bs is 1, the image that receives is carried out weak filtering; When the blocking effect value is 2, the image that receives is carried out strong filtering.

In the process of carrying out strong filtering and weak filtering, present embodiment also can adopt embodiment 1 described method, but, in order further to reduce the complexity and the amount of calculation of filtering algorithm, described strong filtering of present embodiment and weak filtering method are compared with embodiment 1 described method, in filtering, less window width and number of taps have been used.

Also with reference to Fig. 1, be example equally with the horizontal direction below, describe described weak filtering of present embodiment and strong filtering method in detail.

Being filtered into the filter window width a little less than present embodiment is described is 4, and tap number is 3 or 4 difference filtering, and described difference filtering will be recomputated the value of adjacent isles boundary pixel P1, P0, Q0 and Q1, and computational methods are as follows:

At first, pixel P2, P1, P0, Q0, Q1 and Q2 according to both sides, adjacent isles border calculate a small amount of Delta-P1, Delta-P0, Delta-Q0 and the Delta-Q1 that adjusts described pixel P1, P0, Q0 and Q1 respectively, wherein, described being calculated as uses weight coefficient (2 ,-3,1) to calculate Delta-P1, Delta-Q1 in a small amount, the weight coefficient (1 that uses,-2,2 ,-1) calculates Delta-P0, Delta-Q0 in a small amount.Specifically, the computing formula of described a small amount of Delta-P1, Delta-P0, Delta-Q0 and Delta-Q1 is as follows:

$\mathrm{Delta}-P1=\frac{(P2-P1)\×2+(Q0-P1)}{4};$

$\mathrm{Delta}-P0=\frac{(Q0-P0)\×2+(P1-Q1)}{4};$

$\mathrm{Delta}-Q0=\frac{(P0-Q0)\×2+(Q1-P1)}{4};$

$\mathrm{Delta}-Q1=\frac{(Q2-Q1)\×2+(P0-Q1)}{4}.$

Then, arithmetic mean value according to the quantization parameter QP of adjacent two fritters, in the pixel increase and decrease difference threshold table Clip2 relevant that sets up in advance, search corresponding pixel increase and decrease difference threshold clip2 with quantization parameter QP, it also is the thresholding of described a small amount of, search in the pixel increase and decrease difference threshold scope [clip2, clip2] that obtains being limited in a small amount.

At last, again the value of pixel P1, P0, Q0 and Q1 is sued for peace respectively with a small amount of Delta-P1, the Delta-P0, Delta-Q0 and the Delta-Q1 that calculate and limit through pixel increase and decrease difference threshold clip2, obtain the pixel of P1, P0 after the final filtering, Q0 and Q1.

The described filter window width that is filtered into by force is 4, and tap number is 3 mean filter, in filtering, uses weight coefficient (1,2,1) to recomputate pixel P1, P0, Q0 and Q1.Specifically, described mean filter will adopt following formula to recomputate the value of adjacent isles boundary pixel P1, P0, Q0 and Q1:

$P1=\frac{P0+2\×P1+Q0}{4};$

$P0=\frac{P1+2\×P0+Q0}{4};$

$Q0=\frac{Q1+2\×Q0+P0}{4};$

$Q1=\frac{Q0+2\×Q1+P0}{4}.$

By the aforementioned calculation formula, just can obtain pixel value, thereby remove the blocking effect in the video image through both sides, adjacent isles border behind difference or the mean filter.

Difference of the present invention or mean filter method can also be passed through following window width decision method, the filter window width in above-mentioned difference or the mean filter method are decreased to 2 or 3 under certain condition, further reduce amount of calculation.Described window width decision method is specific as follows:

At first, calculate the near pixel in two back gauge borders second, adjacent isles border, as P1 among Fig. 1 and Q1, the absolute value of both sides pixel value difference promptly calculates the absolute value of P2 and P0 pixel value difference, is designated as C4; Calculate the absolute value of Q2 and Q0 pixel value difference, be designated as C5;

Then, with the pixel difference absolute value C4 that calculates and C5 respectively with corresponding piece in pixel value difference thresholding β compare, if C4 greater than β, does not then need pixel P1 is carried out filtering; If C5 greater than β, does not then need pixel Q1 is carried out filtering.Thereby, can be further the window width of mean filter be reduced to 2 or 3 by said method, to reduce the amount of calculation of difference or mean filter.

Equally, because the fritter of image comprises horizontal boundary and vertical boundary, therefore, the filtering of video image also is divided into horizontal filtering and vertical filtering both direction.Though above be the filtering method that example has illustrated adjacent isles with the horizontal direction only, the method for vertical direction filtering is also basic identical.

This shows,, can select the filtering method of different dynamics that the image of importing is carried out The disposal of gentle filter according to blocking effect value between adjacent two fritters by above-mentioned filtering judgement and filtering execution in step.Simultaneously, because method of the present invention compares with embodiment 1 described method, the implementation dynamics of smothing filtering is reduced to 3 grades, and the tap number in the filtering implementation is reduced to 3 or 4, reduced the complexity of filtering algorithm greatly, and amount of calculation.

Specifically, consider, in test, find that the image of the smooth filtering method output that provides with embodiment 1 is compared, do not have very big difference through the image subjective quality of the described filtering method output of present embodiment from the angle of output video image quality; And its objective quality, i.e. the image of the important indicator PSNR of image encoding and existing smoothing filter output is also basic identical, and the bit number that obtains behind the encoding video pictures slightly reduces.

Consider from the angle of amount of calculation, the operand of filtering algorithm mainly is summed up as the operand of add operation and the operand of displacement computing, the described method of present embodiment is compared with embodiment 1 described method, the implementation dynamics of smothing filtering is reduced to 3 grades, and the tap number in the filtering implementation is reduced to 3, make the operand of addition reduce about 13% to 60%, the amount of calculation of displacement algorithm has reduced about 17% to 40%, thereby, make overall amount of calculation reduce about 25% to 40%.

Consider from the angle of arithmetic speed, in the test process of real-time coding, find, adopting the described method of present embodiment to carry out deblocking effect handles, under other central processing unit of Pentium IV level (CPU) computing capability condition, the image sequence of coding QCIF form, its average coding result is than fast 10 to 20 frames of embodiment 1 described smooth filtering method; The image sequence of coding CIF form, its average coding result is wanted fast 5 to 10 frames.Like this, make code rate accelerate greatly.

Consider from hard-wired angle, when using Digital Signal Processing (DSP) device to realize the described filtering method of present embodiment, can be when calculating new pixel, use less number of taps, thereby, make in the computational process that the required memory headroom of opening up significantly reduces in DSP, saved hardware resource, make the easier realization on hardware of the described method of present embodiment.

In order to realize goal of the invention of the present invention, remove the blocking effect in the video image, the present invention also provides a kind of loop filter of realizing above-mentioned filtering method.

Fig. 2 has shown the structure of described loop filter.As shown in Figure 2, loop filter of the present invention mainly comprises: filtering decision unit 100, be used for the image of input is carried out rim detection, judge the true edge that whether has image on the border of adjacent isles, and the blocking effect value of determining two adjacent isles, wherein, different blocking effect values is corresponding one by one with different filtering dynamics; Filtering performance element 200 is used for judging according to the edge detection results of filtering decision unit 100 inputs and definite blocking effect value, for the adjacent isles that has true edge on the border, does not carry out filtering; For the adjacent isles that on the border, does not have true edge,, select the filtering method corresponding to carry out filtering with this filtering dynamics according to the filtering dynamics of its blocking effect value correspondence.

Wherein, described filtering decision unit 100 further comprises: pixel and coded message memory module 110, threshold value determination module 120, pixel value difference computing module 130, blocking effect value computing module 140 and rim detection module 150.

Described pixel and coded message memory module 110 are used to receive and the pixel of the pending image that buffer memory is imported and the coded message of adjacent two fritters, and the pixel and the coded message of buffer memory sent to threshold value determination module 120, pixel value difference computing module 130 and blocking effect value computing module 140.

Wherein, described coded message comprises: affiliated macroblock encoding type, the motion vector of fritter, reference frame and quantization parameter or the like.

Threshold value determination module 120 is according to the arithmetic mean value of the quantization parameter QP of two fritters of input, obtain the average quantization step-length of two fritters, and with the index of this quantization step as pixel difference thresholding table, at described α, search pixel value difference thresholding in interblock pixel value difference thresholding corresponding and the piece in the β table with this quantization step, be α, the β value outputs to rim detection module 150 with these two threshold values then.

Pixel value difference computing module 130 calculates the pixel value difference of pixel on the adjacent isles border according to the pixel that receives, calculative three differences of this module: one is the absolute value of a pair of pixel value difference on both sides, fritter border, promptly described interblock pixel value difference absolute value; Two is the absolute value of two pairs of pixel value differences of fritter inside in addition, promptly described interior pixel value difference absolute value, and difference C1 as previously described, C2 and C3 output to rim detection module 150 with the difference that calculates then.

Rim detection module 150 is used for according to pixel value difference absolute value in the interblock of pixel value difference threshold value in the interblock of thresholding determination module 120 inputs, the piece and 130 inputs of pixel value difference computing module and the piece, adjudicate the true edge that whether has image on this adjacent isles border, when the input interblock pixel value difference absolute value less than interblock pixel value difference thresholding, and the pixel value difference absolute value is less than in the piece during pixel value difference thresholding in the piece, be that C1 is less than α, and C2 is less than β, and C3 is during less than β, then is judged to be not have real edge; Otherwise, then be judged to be and have real edge; Then court verdict is outputed to filtering performance element 200.Need to prove that the method for above-mentioned rim detection is identical with the described method of step 11.

Blocking effect value computing module 140 is used for pixel and the coded message according to pixel and 110 inputs of coded message memory module, for example: information such as the motion vector of affiliated macroblock encoding type, fritter and reference frame are determined the current blocking effect value Bs that handles fritter, and the blocking effect value Bs that determines is outputed to filtering performance element 200.Wherein, definite method of described blocking effect value determines that with step 12 or the described blocking effect value of step 22 method is identical.

Filtering performance element 200 further comprises: filtering trigger module 210, difference filtration module 220 and mean filter module 230.

Described filtering trigger module 210 is according to carrying out the filtering judgement from the edge detection results of rim detection module 150 and from the blocking effect value of blocking effect value computing module 140.If the blocking effect value of adjacent isles is divided into 5 grades, then when the result of calculation that has true edge or blocking effect value on the adjacent isles border is 0, then do not carry out filtering, described filtering trigger module 210 is directly with the pixel and the coded message output that receive; If not having the result of calculation of true edge and blocking effect value on the adjacent isles border is 1 to 3 o'clock, then pixel and the coded message that receives outputed to difference filtration module 220, carry out difference filtering; If not having the result of calculation of true edge and blocking effect value on the adjacent isles border is 4 o'clock, then pixel and the coded message that receives outputed to mean filter module 230, carry out mean filter.And if the blocking effect value of adjacent isles is divided into 3 grades, then when the result of calculation that has true edge or blocking effect value on the adjacent isles border is 0, then do not carry out filtering, described filtering trigger module 210 is directly with the pixel and the coded message output that receive; If not having the result of calculation of true edge and blocking effect value on the adjacent isles border is 1 o'clock, then pixel and the coded message that receives outputed to difference filtration module 220, carry out difference filtering; If not having the result of calculation of true edge and blocking effect value on the adjacent isles border is 2 o'clock, then pixel and the coded message that receives outputed to mean filter module 230, carry out mean filter.

Wherein, the filtering algorithm of difference filter 220 and mean filter 230 can adopt front embodiment 1 or embodiment 2 described differences and mean filter algorithm, therefore, here just has been not described in detail.

Using loop filter of the present invention can select the filtering method of different dynamics that the image of importing is carried out The disposal of gentle filter according to blocking effect value between adjacent two fritters.

Claims (18)

1, a kind of filtering method of removing blocking effect in the video image is characterized in that, described method comprises: two steps are carried out in filtering judgement and filtering, wherein,

Described filtering judgement is: according to the pixel and the coded message that receive image, determine whether to exist on the border of per two adjacent isles the true edge of image and the blocking effect value of determining per two adjacent isles, wherein, different blocking effect values is corresponding one by one with different filtering dynamics;

Described filtering is implemented as: for the adjacent isles that has true edge on the border, do not carry out filtering; For the adjacent isles that on the border, does not have true edge,, select the filtering method corresponding to carry out filtering with this filtering dynamics according to the filtering dynamics of its blocking effect value correspondence.

Whether 2, the method for claim 1 is characterized in that, exist the true edge of image to comprise on the border of described definite two adjacent isles:

A, according to the pixel that receives image, calculate pixel value difference absolute value in the piece of each fritter in the interblock pixel value difference absolute value of adjacent isles and this adjacent isles respectively;

B, pixel value difference thresholding in pixel value difference absolute value and predefined interblock pixel value difference thresholding and the piece in the interblock pixel value difference absolute value that calculates and the piece is compared, if described interblock pixel value difference absolute value is greater than described interblock pixel value difference thresholding, then there is the true edge of image in the interior pixel value difference absolute value of piece that a fritter is arranged in the perhaps described adjacent isles greater than described interior pixel value difference thresholding on the border of two fritters; Otherwise, do not have the true edge of image on the border of two adjacent isles.

3, the method for claim 1 is characterized in that, the blocking effect value of described definite two adjacent isles is: the coded message that receives image is judged,

When any fritter in described two adjacent isles is the macroblock edges fritter of intraframe coding, determine the blocking effect value of blocking effect value for corresponding strong filtering;

When the macroblock edges fritter that described two adjacent isles all are not intraframe codings, when still having the fritter of motion prediction difference or code coefficient difference, determine the blocking effect value of blocking effect value for corresponding weak filtering;

When above condition does not satisfy, determine the blocking effect value of blocking effect value for corresponding not filtering.

4, method as claimed in claim 3 is characterized in that, the macroblock edges fritter of described intraframe coding is to belong to the infra-frame prediction macro block and its border is the fritter of affiliated infra-frame prediction macroblock boundaries;

Described two adjacent isles all are not the macroblock edges fritters of intraframe coding, but having the fritter of motion prediction difference or code coefficient difference is the fritter that satisfies one of following four conditions:

Belong to intra-frame prediction block, but its border not the border of affiliated macro block;

Code coefficient is not zero entirely;

The reference frame that is adjacent fritter is inequality;

The reference frame that is adjacent fritter is identical, and still the difference with the adjacent isles motion vector has one-component at least greater than a pixel.

5, method as claimed in claim 3 is characterized in that, the blocking effect value of the weak filtering of described correspondence further is included as 3 sub-grades;

Definite method of the sub-grade of blocking effect value of the weak filtering of described correspondence is as follows:

Any fritter in described two adjacent isles belongs to the infra-frame prediction macro block, but the little block boundary that belongs to the infra-frame prediction macro block determines that the sub-grade of blocking effect value of corresponding weak filtering is the first sub-grade of weak filtering when not being the border of macro block under it;

When the code coefficient that has a fritter in described two adjacent isles at least is not zero entirely, determine that the sub-grade of blocking effect value of corresponding weak filtering is the second sub-grade;

Reference frames inequality when the reference frame of described two adjacent isles or described two adjacent isles are identical, when but the difference of described two little block motion vectors has one-component greater than a pixel at least, determine that the sub-grade of blocking effect value of corresponding weak filtering is the 3rd sub-grade.

6, the method for claim 1 is characterized in that, described filtering dynamics according to its blocking effect value correspondence selects the filtering method corresponding with this filtering dynamics to be filtered into:

When the blocking effect value is the blocking effect value of corresponding not filtering, do not carry out filtering, directly the image and the coded message of output reception;

When the blocking effect value is the blocking effect value of corresponding weak filtering, the image that receives is all carried out the difference filtering corresponding with weak filtering dynamics;

When the blocking effect value is the blocking effect value of corresponding strong filtering, the image that receives is carried out and the strong corresponding mean filter of filtering dynamics.

7, method as claimed in claim 6 is characterized in that, the method for described difference filtering is as follows:

A1, calculate according to the pixel on both sides, adjacent isles border each pixel in the filter window is adjusted required a small amount of;

B1, according to the arithmetic mean value of the quantization parameter of adjacent two fritters, in the pixel increase and decrease difference threshold table of setting up in advance relevant, search corresponding pixel increase and decrease difference threshold, and a small amount of that step a1 calculates is limited in searches the pixel that obtains and increase and decrease in the difference threshold scope with quantization parameter;

C1, with each pixel in the filter window and its corresponding addition in a small amount through step b1 restriction, obtain the pixel behind interior each pixel filter of filter window.

8, method as claimed in claim 7 is characterized in that, establishes P1, P0 and Q0, Q1 is a tactic pixel on two adjacent isles borders, and wherein, P0 and Q0 are pixel in the described filter window, and a small amount of of its correspondence is Delta-P0, Delta-Q0;

Described being calculated as of step a uses weight coefficient (1 ,-2,2 ,-1) to calculate Delta-P0 and Delta-Q0 in a small amount.

9, method as claimed in claim 8 is characterized in that, establishes P2, Q2 and be on two adjacent isles borders and P1 or Q1 adjacent pixels, and P1 and Q1 also are pixel in the described filter window, and a small amount of of its correspondence is Delta-P1, Delta-Q1;

Described being calculated as of step a uses weight coefficient (2 ,-3,1) to calculate Delta-P1 and Delta-Q1 in a small amount.

10, method as claimed in claim 6 is characterized in that, described difference filtering method is as follows:

A2, calculate according to the pixel on both sides, adjacent isles border each pixel in the filter window is adjusted required a small amount of;

B2, according to the sub-grade of blocking effect value of the corresponding weak filtering of adjacent two fritters, in pixel increase and decrease difference threshold table that set up in advance and the sub-rank correlation of blocking effect value, searching corresponding pixel increase and decrease difference threshold, and a small amount of that step a2 calculates is limited in searching in the pixel increase and decrease difference threshold scope that obtains;

C2, with each pixel in the filter window and its corresponding addition in a small amount through step b2 restriction, obtain the pixel behind interior each pixel filter of filter window.

11, as claim 7 or 10 described methods, it is characterized in that, if P2, P1, P0 and Q0, Q1, Q2 are tactic pixel on two adjacent isles borders, wherein, P1, P0 and Q0, Q1 are pixel in the described filter window, and a small amount of of its correspondence is Delta-P0, Delta-P1, Delta-Q0 and Delta-Q1;

Step a1 is described to be calculated as: use weight coefficient (1 ,-4,4 ,-1) to calculate Delta-P0, Delta-Q0 in a small amount, use weight coefficient (2 ,-4,1,1) to calculate Delta-P1, Delta-Q1 in a small amount.

12, method as claimed in claim 6 is characterized in that, described mean filter is: the pixel according to both sides, adjacent isles border recomputates each point pixel in the filter window.

13, method as claimed in claim 12 is characterized in that, establishes P1, P0 and Q0, Q1 is a tactic pixel on two adjacent isles borders, and wherein, P0 and Q0 are pixel in the described filter window;

Described being calculated as: use weight coefficient (1,2,1) calculating pixel P0 and Q0.

14, method as claimed in claim 13 is characterized in that, establishes P1 and Q1 and also is pixel in the described filter window;

Described being calculated as: use weight coefficient (1,2,1) calculating pixel P1 and Q1.

15, method as claimed in claim 12 is characterized in that, establishes P3, P2, P1, P0 and Q0, Q1, Q2, P3 is a tactic pixel on two adjacent isles borders, and wherein, P2, P1, P0 and Q0, Q1, Q2 are pixel in the described filter window;

Described being calculated as: use weight coefficient (1,2,2,2,1) calculating pixel P0, Q0, use weight coefficient (1,1,1,1) calculating pixel P1, Q1, use weight coefficient (2,3,1,1,1) calculating pixel P2, Q2.

16, a kind of loop filter of removing blocking effect in the video image is characterized in that, comprising:

Filtering decision unit (100) is used for the image of input is carried out rim detection, and the blocking effect value of determining two adjacent isles, and wherein, different blocking effect values is corresponding one by one with different filtering dynamics;

Filtering performance element (200) is used for judging according to the edge detection results of filtering decision unit (100) input and definite blocking effect value, for the adjacent isles that has true edge on the border, does not carry out filtering; For the adjacent isles that on the border, does not have true edge,, select the filtering method corresponding to carry out filtering with this filtering dynamics according to the filtering dynamics of its blocking effect value correspondence.

17, loop filter as claimed in claim 16 is characterized in that, described filtering decision unit (100) further comprises:

Pixel and coded message memory module (110) are used to receive and the pixel of the pending image that buffer memory is imported and the coded message of adjacent two fritters, and output;

Threshold value determination module (120), be used for according to quantization parameter from pixel and two fritters of coded message memory module (110) coded message, be identified for carrying out the interblock pixel value difference thresholding and the interior pixel value difference thresholding of piece of rim detection, and these two threshold values are outputed to rim detection module (150);

Pixel value difference computing module (130) is used for according to the pixel value difference absolute value that calculates pixel on the adjacent isles border from the pixel of pixel and coded message memory module (110), and the absolute difference that calculates is outputed to rim detection module (150);

Rim detection module (150) is used for threshold value and pixel value difference absolute value according to input, adjudicates the true edge that whether has image on the described adjacent isles border, and court verdict is outputed to filtering performance element (200);

Blocking effect value computing module (140), be used for according to from the pixel of pixel and coded message memory module (110) and the blocking effect value that coded message is calculated described adjacent isles, and the blocking effect value determined and described pixel and coded message are outputed to filtering performance element (200).

18, loop filter as claimed in claim 16 is characterized in that, described filtering performance element (200) further comprises:

Filtering trigger module (210), be used for carrying out the filtering judgement according to edge detection results and blocking effect value from described filtering decision unit (100), to the filtering dynamics that has image true edge or blocking effect value correspondence on the adjacent isles border is the fritter of not filtering, and the pixel and the coded message of input are directly exported; Filtering dynamics to blocking effect value correspondence is the fritter of weak filtering, and pixel and the coded message of importing outputed to difference filtration module (220); Filtering dynamics to blocking effect value correspondence is the fritter of strong filtering, with the pixel and the coded message output mean filter module (230) of input;

Difference filtration module (220) is used for the pixel and the coded message of input are carried out difference filtering;

Mean filter module (230) is used for the pixel and the coded message of input are carried out mean filter.