一种基于立体视觉显著性的多视点视频比特分配方法

A bit allocation method for multi-view video coding based on stereoscopic visual saliency

针对多视点立体视频压缩编码,提出了一种基于立体视觉显著性的比特分配方法。研究综合利用多视点立体视频数据中场景的运动、深度以及深度边缘信息提取人眼感兴趣区域(ROI)的方法;然后根据ROI的划分结果优化区域比特分配。实验结果表明,本文提出的算法能有效提高ROI区域的编码性能,同时整体视频的率失真性能有一定程度的提高。

Human visual system （HVS） based video coding has obvious research significance. Saliency model has been employed to the detection of interesting regions and thus applied to the region based bit allocation of video coding. The regions of interest （ROIs） are allocated more bits and other regions are allocated fewer bits in order to reduce the transmission bits while keeping good perceptual quality. How- ever, the conventional ROI based bit allocation algorithms in 2D video coding cannot be applied directly to 3D video coding. This paper proposes a bit allocation method based on stereoscopic visual saliency for multi-view video coding （MVC）. We propose an ROI segmentation model which utilizes the information of motion, depth and the edge of depth maps. The extracted results show that our proposed segmentation model can achieve good performance. Then, a regional bit allocation scheme is realized by adjusting quan- titative parameter （QP） based on the ROI segmentation results to allocate more bits to ROI and fewer bits to other regions. Video sequences with different characteristics are utilized to evaluate the perform- ance of our proposed bit allocation method. Experimental results demonstrate that the proposed method can effectively enhance the performance of ROI as well as the whole video.