CN1207915C - Half pixel filtering method for video compression domain - Google Patents

Abstract

The present invention relates to a video half-pixel filtering method of a compressed domain, which can enhance video quality and reduce the computation complexity of video operation. The present invention is characterized in that an adopted half-pixel filtering method is selected according to the position of an extracting block; the half-pixel filtering method is used for filtering coefficients of the compressed domain to obtain a coefficient of the compressed domain corresponding to a half pixel position. The present invention acts in the compressed domain, filtering can be carried out to the coefficients of the compressed domain, and the present invention is independent of video compressing standards.

Description

Video compression territory half-pix filtering method

Technical field

The present invention relates to the technical field of network, the method for particularly a kind of video compression territory half-pix filtering.

Technical background

Along with the development of network technology, computing technique, based on network Video Applications occurs in a large number.There is isomerism in the equipment of access network, and computing capability there are differences, and the video standard of support also has nothing in common with each other.The video coding technique computation complexity is bigger, can't satisfy the demand of all types of user.Video code conversion, editing technique can be operated compressed video code flow again, improve video processing speed.

So-called half-pix is meant at the pixel between whole pixel.In discrete picture was handled, half-pix point was not that physics exists, and is to exist in order to satisfy certain purpose, for example accurately motion compensation, picture up-sampling etc.

Video encoding standard is generally supported the half-pix estimation, and the half-pix estimation has improved video quality greatly, has reduced the code stream code check.If the compressed domain video treatment technology is not supported half-pix, can't satisfy current demand.

Video encoding standard extensively adopts dct transform, can effectively reduce the spatial redundancies of video.So related compression domain is meant the result of discrete cosine transform dct transform among the present invention.

Summary of the invention

The object of the present invention is to provide the method for a kind of video compression territory half-pix filtering.

A kind of compressed domain video half-pix filtering method can improve video quality, reduces the computation complexity of vision operation, and its step is as follows:

1) the half-pix filtering method that adopts according to the choice of location of extracting piece;

2) with this half-pix filtering method to the filtering of compression domain coefficient, obtain corresponding half-pixel position compression domain coefficient.

Select the half-pix filtering method of employing according to the requirement of compressed domain video half-pix filtering.The method of a kind of video compression territory half-pix filtering, its step is as follows:

1) according to extracting piece

The position, determine

Same B ₁Relevant line number m, columns n; Line number m,

Columns n equals 0 or positive integer;

2), determine the filtering method that will adopt according to line number m, columns n;

3) according to extracting piece With the correlation circumstance of adjacent block, determine related blocks;

4) use filtering method that related blocks is carried out filtering, obtain

Description of drawings:

Fig. 1 is an anti-motion compensation schematic diagram of the present invention.

Embodiment

Compressed domain video half-pix filtering problem as shown in Figure 1, known compression blocks B _i, i ∈ [1,2,3,4] will extract compression blocks from certain half-pixel position

Obviously, can't directly extract compression blocks Need extract compression blocks simply, fast by certain algorithm The present invention proposes a kind of half-pix filtering algorithm, promptly passes through B _iProduct, summation operation with the half-pix filtering method can achieve the above object, and solve the half-pix filtering problem.The concrete steps of half-pix filtering are:

1) according to extracting piece

The position, determine Same B ₁Relevant line number m, columns n; Line number m, columns n equal 0 or positive integer;

2), determine the filtering method that will adopt according to line number m, columns n;

3) according to extracting piece With the correlation circumstance of adjacent block, determine related blocks;

4) use filtering method that related blocks is carried out filtering, obtain If E ^s=1,2 ..., s}, wherein s is an integer, and we at first define 1≤s≤8. two classes s * s and tie up sparse matrix, and dimension is meant the dimension of matrix, $R_s={\left(R_s(i,j)\right)}_{i,j\∈E^s}$ and $L_s={\left(L_s(i,j)\right)}_{i,j\∈E^s},$ Wherein

The definition matrix:

$P_m=\left[\begin{array}{cc}0& R_m\\ 0& 0\end{array}\right]$

$P_n=\left[\begin{array}{cc}0& 0\\ L_n& 0\end{array}\right]$

P _m, P _nCorresponding dct transform result is

H _m＝DCT(P _m)

H _n＝DCT(P _n)

H _m, H _nBe compression domain half-pix filtering method, according to extracting piece These two filtering methods of position grouping, respectively to B _i, i=1,2,3,4 carry out filtering, and with the filtering results added, just obtain The compression domain coefficient.

The meaning that we at first define two classes s * s dimension sparse matrix is definition two matroids, and this dimension is meant the dimension of matrix;

Rs and Ls matrix are used for defining matrix Pm and Qn again; And Pm and Qn are actually used in the matrix that filtering is used;

Rs, Ls, Pm and Qn are the matrixes that oneself defines, the coefficient matrix that adopts when being filtering;

Relevant line number and columns in m and the objective matrix that n equals to extract, m=2 for example, n=3 represents to extract the matrix of one 2 row 3 row;

I is the serial number that will extract piece, first piece i=1 of Ti Quing for example, second piece i=2 of extraction.

Compressed domain video half-pix filtering method of the present invention can improve video quality, reduces the computation complexity of vision operation.Feature of the present invention is: according to the half-pix filtering method of the choice of location employing of extracting piece; To the filtering of compression domain coefficient, obtain corresponding half-pixel position compression domain coefficient with this half-pix filtering method.The present invention acts on compression domain, can carry out filtering to the compression domain coefficient, and not rely on video compression standard.

Claims (4)

1. a compressed domain video half-pix filtering method can improve video quality, reduces the computation complexity of vision operation, and its step is as follows:

1) the half-pix filtering method that adopts according to the choice of location of extracting piece;

2) use the half-pix filtering method to the filtering of compression domain coefficient, obtain corresponding half-pixel position compression domain coefficient.

2. by the described compressed domain video half-pix of claim 1 filtering method, it is characterized in that: the half-pix filtering method of selecting employing according to the requirement of compressed domain video half-pix filtering.

3. the method for a video compression territory half-pix filtering, its step is as follows:

1) according to extracting piece The position, determine

Same B ₁Relevant line number m, columns n; Line number m, columns n equal 0 or positive integer;

2), determine the filtering method that will adopt according to line number m, columns n;

3) according to extracting piece With the correlation circumstance of adjacent block, determine related blocks;

4) use filtering method that related blocks is carried out filtering, obtain

4. by the method for the described video compression of claim 3 territory half-pix filtering, it is characterized in that: establish E ^s=1,2 ..., s}, wherein s is an integer, 1≤s≤8, we at first define two classes s * s and tie up sparse matrix, and dimension is meant the dimension of matrix, $R_s={\left(R_s(i,j)\right)}_{i,j\∈E^s}$ and $L_s={\left(L_s(i,j)\right)}_{i,j\∈E^s},$ Wherein

The definition matrix:

$P_m=\left[\begin{array}{cc}0& R_m\\ 0& 0\end{array}\right]$

$P_n=\left[\begin{array}{cc}0& 0\\ L_n& 0\end{array}\right]$

P _m, P _nCorresponding dct transform result is

H _m＝DCT(P _m)

H _n＝DCT(P _n)

Pm and Qn are actually used in the matrix that filtering is used;

The coefficient matrix that Rs, Ls, Pm and Qn adopt when being filtering;

Relevant line number and columns in m and the objective matrix that n equals to extract,

I is the serial number that will extract piece,

H _m, H _nBe compression domain half-pix filtering method, according to extracting piece These two filtering methods of position grouping, respectively to B _i, i=1,2,3,4 carry out filtering, and with the filtering results added, just obtain The compression domain coefficient.

Images