一种深度图的时域下采样编码及重建方法  

Temporal subsampling based depth maps coding and the reconstruction method

作　　者：葛川[1] 刘琚[1,2] 元辉[1,3] 肖依凡 李凤荣[3] 

机构地区：[1]山东大学信息科学与工程学院,山东济南250101 [2]山东大学苏州研究院,江苏苏州215123 [3]中国科学院上海微系统与信息技术研究所无线传感网与通信重点实验室,上海200050

出　　处：《西安电子科技大学学报》2016年第4期160-165,171,共7页Journal of Xidian University

基　　金：国家自然科学基金资助项目(61201211;61571274);教育部博士点基金资助项目(20120131120032;20130131110029);山东省优秀中青年科学家奖励基金资助项目(BS2012DX021);中国科学院无线传感网与通信重点实验室开放课题资助项目(2013002);山东大学青年学者未来计划资助项目(2015WLJH39);苏州市科技计划资助项目(SYG201443)

摘　　要：为提高三维视频系统的编码效率,提出了一种针对中间视点的深度时域下采样编码方法.首先确定深度图像的丢弃方式;然后利用基于时间一致性和视点间的相关性,恢复出丢弃的深度图像,并对恢复出的结果进行维纳滤波,进一步提高恢复的深度图的质量.该方法需要向编码端发送维纳滤波器系数和重建深度选择标识符.实验结果表明,在保证同等码率的前提下,解码端重建的虚拟视图的峰值信噪比最大能提升0.229dB,平均提升0.130dB,编码性能得到改善.In order to improve the coding efficiency of a three-dimensional video system, a depth temporal subsampling based coding method for the intermediate view is proposed. In this paper, the manner of how to discard depth frames is firstly determined. Then, those discarded depths are reconstructed utilizing temporal consistency and multiview correspondences. Finally, in order to further improve the quality of the reconstructed depth, the wiener filter is implemented on the reconstructed results. Filter coefficients and indicators for selecting the depth predictor should be transmitted to the decoder. Experimental results demonstrate that a maximum 0. 229 dB and average 0. 130 dB peak signal-to noise ratio （PSNR） gain could be achieved for the virtual view reconstructed by the decoder while maintaining the same coding bit rate, and that the coding efficiency can be improved.

关 键 词：视频编码 立体图像处理 深度图时域下采样 时间一致性 视点间相关性 

分 类 号：TN919.8[电子电信—通信与信息系统]