视频编码中一种新的用于几何块划分的变换结构

A New Transform Structure for Geometry Motion Partitioning in Video Coding

针对几何块划分技术提出了一种任意形状的变换结构.虽然几何块划分技术比现有视频编码标准中采用的块划分方式更为灵活,但其所对应的变换结构并未得到合理设计,从而导致变换效率下降.因此,提出对两个任意形状的残差块分别进行任意形状的变换,并通过翻转的方法保留像素间的相关性.实验结果表明,将任意形状的变化结构应用于几何块划分技术可以节约1.16%～3.05%比特率.

This paper proposes a transform structure with arbitrary shape blocks for geometry block partitioning. A square coding block is split into two arbitrary shaped prediction blocks by geometry partitioning,which is more flexible than the partitioning method used in current video coding standards.In the current design, using a square transform block for combined square residual block is not reasonable and will affect transform efficiency.To further improve the transform structure,we propose that the two arbitrary shaped residual blocks are transformed separately with arbitrary shaped transform blocks.A turnover method is used to retain the correlation of residuals.Experimental results show that a bit rate saving of 1.16% ～3.05% can be achieved.