基于改进期望最大化算法的分布式视频编码

目前,分布式视频编码(DVC)由于具有低复杂编码特性而成为视频编码领域的研究热点。DVC中,边信息性能对系统性能影响很大,一般而言,边信息性能越好,则整个系统压缩性能越好。本文研究了一种基于贝叶斯准则的期望最大化(Expectation Maximization,EM)边信息产生方法,并从运动搜索模板和初始概率模型两方面对EM算法提出了改进。实验表明,在几乎相同的率失真性能下,两种改进算法的学习时间分别缩短了30%和16%。

Distributed Video Coding Based on Improved Expectation Maximization Algorithm

Nowadays,the distributed video coding(DVC)has been receiving more and more attentions in video coding field due to its property of low-complexity encoding,which makes it suitable for low-power video capturers.In DVC,side information is very important.Generally,the better side information,the better compression performance of the whole system.In this paper,firstly,the optimal method to side information with Bayes-criterion-based expectation maximization(EM)algorithm is studied;then two modified methods are proposed to improve EM from the points of its motion-search model and initial probability model.Finally,experimental results show that the proposed two methods can save about 30% and 16% in learning time respectively,while maintaining the almost same rate-distortion performances.