面向绘制质量的深度图像快速帧内编码

Fast Depth Map Intra Mode Coding Focus on Quality of View Synthesis

与彩色视频用来直接显示不同,在三维视频系统中深度视频序列的作用是在绘制虚拟视点时提供所需的几何信息,所以直接将现有编码算法应用于深度图像存在一定的局限性。针对深度视频序列的作用以及深度图像自身特征,提出了一种面向绘制质量的深度图像快速帧内编码方法,该方法包括基于深度图像统计特性和空域相关性的图像区域划分算法、HEVC的快速编码单元(CU)和预测块(PB)决策算法和帧内编码模式预先选择算法。实验结果表明:与直接利用HEVC测试软件编码深度视频相比,该快速算法在保证几乎相同的主客观绘制质量的前提下,每个视点的编码速度平均提升了35%以上。

The depth map provides the geometrical information for virtual view rendering in 3D video system and it is not displayed in screen, which is different from the color video. As a result, it is necessary to improve the existing compression algorithm when compress the depth map. Aiming at the function and feature of depth map, this paper proposed a fast depth map intra mode compression method focus on quality of view synthesis. This fast method include image region segmentation algorithm based on the statistic characteristic of depth map and the spatial correlation, fast CU and PB decision strategy in HEVC and fast intra mode decision algorithm. The experimental results show that the proposed fast method can reduce computation time more than 35% and maintain almost the similar quality of synthesized view.