基于3D混合树和视觉特性的视频可分级编码算法  

Scalable video coding algorithm based on 3D hybrid tree and visual characteristics

作　　者：付明哲[1] 王相海[1,2] 宋传鸣[1] 

机构地区：[1]辽宁师范大学计算机与信息技术学院,辽宁大连116029 [2]智能计算与信息处理教育部重点实验室,湖南湘潭411105

出　　处：《通信学报》2012年第11期100-107,共8页Journal on Communications

基　　金：国家自然科学基金资助项目(41271422);辽宁省自然科学基金资助项目(20102123);辽宁省博士科研启动基金资助项目(20121076);计算机软件新技术国家重点实验室开放基金资助项目(KFKT2011B09;KFKT2011B11);南京邮电大学图像处理与图像通信江苏省重点实验室开放基金资助项目(LBEK2010003;LBEK2011001);苏州大学江苏省计算机信息处理技术重点实验室开放课题基金资助项目(KJS1124);智能计算与信息处理教育部重点实验室(湘潭大学)开放课题基金资助项目(2011ICIP06)~~

摘　　要：分析了视频数据的3D小波系数分布特性,提出了一种基于混合3D树型结构和HVS特性的视频可分级编码算法。首先,依据小波低、高频系数的自相关性,确定相应的树型结构来扫描和处理时间维上的低、高频系数,明显减少了用于定位重要系数的同步信息;其次,依据人类视觉系统对各频率子带敏感程度的不同,对各子带系数进行加权,使得重构视频的重要系数得以排在码流前端,从而在很大程度上提高了中低码率下视频的重构质量。对多种标准测试视频的仿真实验验证了本文算法的有效性,与非对称树型结构编码方案和单一时空方向树结构方案相比,该算法解码图像的Y、U和V 3个分量的均峰值信噪比分别高出0.65dB、1.75dB、1.77dB和0.23dB、2.11dB、1.72dB。此外,算法有效抑制了振铃效应,并获得了更好的主观效果。The distribution characteristic of three-dimensional wavelet coefficients of video data was analyzed, and a scalable video coding algorithm was subsequently addressed based on hybrid three-dimensional tree and human visual system （HVS） characteristics. First, the hybrid tree structure was adaptively determined according to the auto-correlation of low-pass and high-pass coefficients. It reduced obviously the number of synchronization bits locating significant wavelet coefficients when scanning and processing low-pass and high-pass coefficients in temporal dimension. Second, each wavelet coefficient was weighted in terms of HVS sensitivity to its corresponding subband. Significant coefficients thus tended to be coded with high priority and arranged at the front of bitstream, and the reconstructed video quality was improved at low and medium bitrates to a great extent. Experimental results in terms of peak signal-to-noise ratio （PSNR） verified the effectiveness of the proposed algorithm on several test videos with varying characteristics. 0.65dB, 1.75dB, and 1.77dB higher PSNR are gained than asymmetric 3-D orientation tree for Y, U, and V components, respectively. Moreover, 0.23dB, 2.1 ldB, and 1.72dB higher PSNR are reached than single temporal-spatial orientation tree separately for Y, U, and V components. Besides, better subjective quality is obtained through effectively attenuating ringing artifact.

关 键 词：视频编码 混合树结构 3D小波变换 零树 人类视觉系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]