基于模式相关性的多视点视频编码宏块模式快速选择算法  

Fast Macroblock Mode Decision Algorithm for Multiview Video Coding Based on Mode Correlation

作　　者：彭宗举[1,2,3] 蒋刚毅[1,2] 郁梅[2,4] 

机构地区：[1]中国科学院计算技术研究所,北京100080 [2]宁波大学信息科学与工程学院,浙江宁波315211 [3]中国科学院研究生院,北京100080 [4]南京大学计算机软件新技术国家重点实验室,江苏南京210093

出　　处：《光学学报》2009年第5期1216-1222,共7页Acta Optica Sinica

基　　金：国家自然科学基金(60672073、60832003、60872094);教育部博士点基金(200816460003);浙江省教育厅项目(20070954);宁波市自然科学丛金(2007A610037)资助课题

摘　　要：针对多视点视频编码计算量大的问题,提出了一种多视点视频编码快速宏块模式选择算法。详细分析了多视点视频编码校验模型的宏块模式选择结果的分布特性以及率失真特性,提出了宏块模式相关性和模式聚集度的概念。对测试序列的宏块模式相关性分析表明当前宏块的最优模式和大多数相邻宏块的最优模式相同。从相邻宏块区域聚集度的角度分析了当前宏块模式预测准确度。在算法实现中采用相邻宏块的率失真代价均值形成的阈值条件决定是否进行全搜索。实验结果表明该算法在保持编码率失真性能的基础上,能提高多视点视频编码的速度2.43～4.08倍。In order to reduce computational complexity of multiview video coding, a fast macroblock mode selection algorithm is proposed. On the basis of a detailed analysis of the distribution and rate-distortion features of macroblock modes in joint multiview video model （JMVM）, conceptions of macroblock mode correlation and mode aggregative degree are put forward. The analysis of the macroblock mode correlation of the test sequence shows that the optimal mode of the current macroblock is the same as that of the most neighboring macroblocks. The precision of the macroblock mode prediction is explored from the angle of the aggregative degree of the neighboring macroblock regions. In the implementation of the fast algorithm, a threshold formed by the average rate distortion cost of the neighboring macroblocks determines the necessity for full searching all macroblock modes. Experimental results demonstrate that the proposed fast algorithm promotes the encoding speed by 2.43～4.08 times in comparison with JMVM, while the rate distortion performance nearly remains same.

关 键 词：图像处理 多视点视频编码 模式选择 模式相关性 快速算法 

分 类 号：TN919.81[电子电信—通信与信息系统]