Intra subblock partitioning (Intra Sub-Partitions, ISP) based on the size of the luminance prediction block is divided into two or four sub-blocks in the horizontal direction or the vertical frame. ISP allows most block size is divided 4x8 or 8x4. If the block is equal to 4x8 or 8x4 is divided into two sub-blocks (VVC provisions must be at least 16 samples per block), if more than 4x8 or 8x4 is divided into four sub-blocks.
| Block size | The number of sub-blocks |
|---|---|
| 4x4 | Not divided |
| 4x8 and 8x4 | 2 |
| other | 4 |
The figure is divided into 4x8 or 8x4 ISP blocks:
The figure is divided ISP greater than 4x8 or 8x4 blocks:
Each sub-block by adding the prediction signal to obtain a residual signal reconstructed signal, wherein the residual signal by entropy decoding, inverse quantization, inverse changes obtained. Each sub-block of the reconstructed signal may be used to construct a prediction value of a next sub-block. The first sub-block containing the pixel sub-blocks in the upper left corner of the CU, each subblock is then treated in accordance with the downward (horizontal division) or right (vertical division). All sub-blocks share the same intra prediction mode.
Here is the ISP coding tools to interact with other cases:
-
When a block using the MRL (multi-line prediction), ISP closed.
-
Entropy encoding coefficients of length: When using ISP, all entropy coding the sub-blocks are 16 samples in one group. This is mainly for the sub-block is smaller than a dimension after impact divided ISP four. Entropy coding coefficient group other cases are 4x4. In the following table.
| Block size | Coefficient group size |
|---|---|
| 1xN, N ≧ 16 | 1x16 |
| Nx1, N ≧ 16 | 16x1 |
| 2xN,N≧8 | 2x8 |
| Nx2,N≧8 | 8x2 |
| other | 4x4 |
-
CBF: default ISP using at least one non-zero CBF subblock. If there are n sub-blocks, before the n-1 is 0, the default is one n-th.
-
MPM: When using the ISP, the default MPM use in the prediction mode, and without the addition of MPM in DC mode. If the ISP is divided horizontally in the MPM mode priority level, vertically divided vertical mode is preferred.
-
Transform kernel size restrictions: DCT-II used when converting a size larger than 16.
-
PDPC: When using the ISP, the default closed PDPC.
-
MTS flag：如果CU使用ISP，则该CU的MTS flag被置为0，且不用传到解码端。因此编码端不需要对每个子划分使用不同的变化进行RD计算。其变换模式会根据其帧内预测模式、处理顺序和块尺寸来进行选择。因此不需要在码流中传输其变换模式。例如对于wxh的子块，其水平变换模式和垂直变换依据下面的规则选择：
-
如果w=1或h=1，不进行水平和垂直变换。
-
如果w=2或w>32，进行DCT-II水平变换
-
如果h=2或h>32，进行DCT-II垂直变换
-
其他情况根据下表进行选择。
-
| 帧内预测模式 | 水平变换模式 | 垂直变换模式 |
|---|---|---|
| planar，角度31，32，34，36，37 | DST-VII | DST-VII |
| DC，角度33，35 | DCT-II | DCT-II |
| 角度2，4，6...28，30，39，41，43...63，65 | DST-VII | DCT-II |
| 角度3，5，7...27，29，38，40，42...,64,66 | DCT-II | DST-VII |
感兴趣的请关注微信公众号Video Coding
