摘要
在立体视频序列编码中 ,左图像流按 MPEG- I/II的视频编码标准进行编码 ,而对右图像流只对参考帧 (I及 P帧 )进行编码 ,右图像流的 B帧不编码及传送 ,依靠在解码端从各个参考帧中进行帧重建及内插获得 .该文中的可变尺寸块分割方法采用基于多分辨率四叉树分解的运动分割 ,对右 B帧的重建提出了按照匹配块和失配块分别处理的方法 ,并根据右 B帧中的块与右 I、P帧之间或与左 B帧之间相关性大小来确定各个块的位置和内容 .该文还提出了一种可避免误匹配和对显露区进行正确填充的针对交叠块的帧估计和内插方法 .仿真实验表明 ,作者提出的图像重建方法与 Monson H.Hayes所提出的基于固定块的方法相比 ,能够使得重建图像的 PSNR增加约 1.2 5d B。
In stereo video sequence coding, the left image stream is coded according to the MPEG I/II video coding standard. In the right image stream, only reference frames (I and P) are coded, the B frames in the right image stream are neither coded nor transmitted at the encoder. They are reconstructed from the reference frames using a frame reconstruction and interpolation technique. The variable block segmentation in this paper is based on a irregular multi resolution quadtree motion segmentation. It is proposed that reconstruction of right B frame is processed as matched blocks and unmatched blocks separately, and the positions and contents of these blocks are decided from correlation between blocks in right B frames and right I, P frames or left B frames. This paper also presents a novel frame reconstruction and interpolation technique to avoid false matching and to fill the exposed regions correctly. The simulation results show that the average PSNR is about 1.25 dB higher than Hayes's method which is based on fixed blocks, and the reconstructed frames have a much better visual quality.
出处
《上海大学学报（自然科学版）》
CAS
CSCD
2001年第6期471-476,共6页
Journal of Shanghai University:Natural Science Edition
基金
国家自然科学基金 ( 6 9972 0 2 7)资助项目
