摘要
通过深度相机或者深度估计软件获取的深度视频准确性差,影响深度视频的编码压缩性能。因此,本文提出了一种面向HEVC的深度视频预处理算法。首先提取深度视频的边缘获取边缘掩膜;其次根据边缘掩膜对深度序列分类进行空域平滑;再次将彩色和空域平滑后的深度视频进行时空转换,并以转换后的彩色视频为辅助对转换后的深度视频进行时域平滑,以增强深度视频的时间相关性;最后对时域平滑处理完的深度视频进行逆时空转换,获取最终的预处理深度视频。实验结果表明,提出的处理方法既可以保证虚拟视点质量基本不变,又能够节省7.00%～23.78%的码率。
The depth video obtained by depth camera or depth estimation software is inaccurate, which will affect the compression performance. So, a HEVC-oriented depth video preprocessing algorithm is proposed. Firstly, the edge-mask of depth video is abstracted. Secondly, the depth pixels are classified and spatially smoothened with the aid of edge-mask. Then, spatial smoothened depth video and its associated texture video are used to perform the Temporal Spatial Transform （TST）, and the transformed depth video is temporally smoothened with the aid of transformed texture vide so as to enhance its temporal correlation. Finally, the temporal smoothened depth video is processed by the Inverse Temporal Spatial Transform （ITST） to get the final preprocessed depth video. The experimental results show that bit rate is saved by the depth video preprocessing algorithm, ranging from 7.00% to 23.78%, while the proposed preprocessing algorithm keeps the rendering quality of virtual view images.
出处
《光电工程》
CAS
CSCD
北大核心
2014年第1期87-94,共8页
Opto-Electronic Engineering
基金
国家自然科学基金(61171163
61071120
61271270
61111140392
U1301257)
宁波市自然科学基金(2012A610041)
关键词
深度视频
边缘掩膜
时空转换
绘制质量
depth video
edge-mask
temporal spatial transform
rendering quality
