Current Picture Referencing mode with Affine Model in VVC
Siqi Cao, Haiyang Han, Fan Liang, Jun Wang, Hongmei Liu · 2019
The next video coding standard Versatile Video Coding (VVC) is developing, and the current picture referencing (CPR) mode for processing Screen Content Video has been adopted in the third version of reference software Video Test Model (VTM3.0). It searches the block vector from the current block to the reference block in the same picture instead of searching the regular motion vector from the current picture to the reference picture. However, it only considers the translational motion compensation, using one vector to represent the movement of the whole block. To fit into more movements in the real world, intra prediction modes with affine motion model based on CPR mode called CPR-Affine mode is proposed in this paper innovatively. And it is the first to apply the affine motion model to intra prediction. First, choose the partial current picture to be the only reference picture, and the reference area must comply with some constraints to ensure that valid reference block can be obtained during decoding. Second, search a set of block-affine vectors (BVAffi) iteratively. BVAffi represents a set of the block vectors of the control points which is the top-left and the top-right corner samples of the current block, and the best BVAffi means the minimum predicted error after compensation. Meanwhile, in order to search the best BVAffi, establish one gradient filter with Sobel operation. The proposed algorithm in this paper is significant to introduce the affine motion model into intra prediction modes. Compared to VTM3.0 with CPR mode turning on, the experimental results demonstrate the efficiency of our proposed algorithm.