Integer transform function for video compression systems
View Patent ↗The present invention is directed to video coding/decoding and discloses a method for transforming to/from transform coefficients and residual pixel data in moving pictures by a set of semi-orthonormal basis vectors. The basis vectors are derived from conventional DCT or KTL matrixes, but relaxes to some extend the requirements for orthogonality, norm equality and element size limitation. In this way the present invention provides improved coding efficiency and lower complexity compared to previously used integer transforms.
1. A method comprising:
transforming, by a processor, a first block of residual pixel values of a moving picture by a transform matrix formed by a number of basis vectors generating a corresponding second block of transform coefficients, the basis vectors being rows of the transform matrix,
wherein the transform matrix is derived from a Discrete Cosine Transform (DCT) or a Karhunen-Loeve Transform (KLT), at least one pair of the basis vectors being non-orthogonal, norms of all the basis vectors have a deviation less than 1% and elements of the basis vectors are less than 32, each norm being a square sum of terms included in only one row, respectively, in the transform matrix, said transform matrix being an integer transform matrix.
2. The method according to claim 1 ,
wherein the basis vectors are used respectively in horizontal and vertical direction of the first block of residual pixel values to generate the corresponding second block of transform coefficients.
3. The method according to claim 1 ,
wherein the blocks represent a first squared area of 8×8 pixels in a picture.
4. The method according to claim 3 ,
wherein said basis vectors are defined as:
[13
13
13
13
13
13
13
13]
[17
16
11
3
−3
−11
−16
−17]
[17
7
−7
−17
−17
−7
7
17]
[16
−3
−17
−11
11
17
3
−16]
[13
−13
−13
13
13
−13
−13
13]
[11
−17
3
16
−16
−3
17
−11]
[7
−17
17
−7
−7
17
−17
7]
[3
−11
16
−17
17
−16
11
−3].
5. A method comprising: transforming, by a processor, a first block of transform coefficients by a transform matrix formed by a number of basis vectors generating a corresponding second block of residual pixel values of a moving picture, the basis vectors being horizontal rows of the transform matrix,
wherein the transform matrix is derived from a Discrete Cosine Transform (DCT) or a Karhunen-Loeve Transform (KLT), at least one pair of the basis vectors being non-orthogonal, norms of all the basis vectors have a deviation less than 1% and elements of the basis vectors are less than 32, each norm being a square sum of terms included in only one row, respectively, in the transform matrix, said matrix being an integer transform matrix.
6. The method according to claim 5 ,
wherein the basis vectors are used respectively in horizontal and vertical direction of the first block of transform coefficients to generate the corresponding second block of residual pixel values.
7. The method according to claim 5 ,
wherein the blocks represent a first squared area of 8×8 pixels in the moving picture.
8. The method according to claim 7 ,
wherein said basis vectors are defined as:
[13
13
13
13
13
13
13
13] T
[17
16
11
3
−3
−11
−16
−17] T
[17
7
−7
−17
−17
−7
7
17] T
[16
−3
−17
−11
11
17
3
−16] T
[13
−13
−13
13
13
−13
−13
13] T
[11
−17
3
16
−16
−3
17
−11] T
[7
−17
17
−7
−7
17
−17
7] T
[3
−11
16
−17
17
−16
11
−3] T .