论文部分内容阅读
多视点彩色加深度(MVD)视频是三维(3D)视频的主流格式。在3D高效视频编码中,深度视频帧内编码具有较高的编码复杂度;深度估计软件获取的深度视频由于不够准确会使深度图平坦区域纹理增加,从而进一步增加帧内编码复杂度。针对以上问题,本文提出了一种联合深度处理的深度视频帧内低复杂度编码算法。首先,在编码前对深度视频进行预处理,减少由于深度图不准确而出现的纹理信息;其次,运用反向传播神经网络(BPNN,backpropagation neural network)预测最大编码单元(LCU,largest coding unit)的最大划分深度;最后联合深度视频的边缘信息及对应的彩色LCU最大划分深度进行CU提前终止划分和快速模式选取。实验结果表明,本文算法在保证虚拟视点质量的前提下,BDBR下降0.33%,深度视频编码时间平均节省50.63%。
Multi-view color depth enhancement (MVD) video is the mainstream format for three-dimensional (3D) video. In 3D high-efficiency video coding, the intra-frame coding of the depth video has high coding complexity; the depth video obtained by the depth estimation software can increase the texture in the flat area of the depth map due to inaccuracy, thereby further increasing the intra-frame coding complexity. In view of the above problems, this paper presents a joint deep processing of intra-frame low complexity video coding algorithm. Firstly, the depth video is preprocessed before encoding to reduce the texture information due to the inaccurate depth map. Secondly, the largest coding unit (LCU) is predicted by backpropagation neural network (BPNN) Finally, the edge information of the joint depth video and the corresponding maximum depth of the color LCU are used to divide the terminal into advanced CUs and select the fast mode. The experimental results show that the proposed algorithm reduces the BDBR by 0.33% and the depth of video coding by 50.63% on the premise of ensuring the quality of virtual viewpoints.