基于深度信息的多视点视频快速模式选择算法

Multi-view Video Fast Mode Selection Algorithm Based on Depth Information

针对多视点视频编码原始模式选择算法编码复杂度极高的问题,分析多视点视频加深度视频中宏块模式与深度信息间的关系,提出一种基于深度信息的多视点视频快速模式选择算法。结合深度值将三维空间分成远、近、中3个区域分别处理。针对复杂度最高的中距离区域,将全局视差矢量指向的参考视点对应块及其周边宏块采用率最高的模式作为候选模式,提前分离出可能采用大块模式的宏块,利用基于运动信息的深度平坦度动态阈值判定方法确定最终的宏块预测模式。实验结果表明,该算法在基本保证率失真性能的情况下,与全搜索算法相比,可平均节省71.70%的搜索点数,明显降低了编码复杂度。

The original Multi-view Video Coding（MVC） mode selection algorithm utilizes full mode search method to find the best prediction mode. It can achieve good performance but also brings the high complexity. This work analyzes the relationship between Macro Block（MB） mode and depth values in Multi-view Video plus Depth（MVD）, and proposes a mode selection algorithm for MVC based on depth information. The depth values are utilized to divide the 3D space into remote, close and midrange areas. Each area is handled respectively. For the midrange, area which is the most complicated one, the most frequently used mode of the corresponding MB and its surrounding MB in reference view are Used as the Most Likely used Mode（MLM） to separate the MB which may use large partition mode. The dynamic depth flatness combined with the motion information method is used to determine the final MB mode. Experimental results show that the proposed algorithm can save 71.70% search points in average compared with full search algorithm, which remarkably reduces the complexity of the mode decision process while maintaining nearly the same rate-distortion performance.