Research On GOP Structure In Distributed Video Coding

Distributed Video Coding(DVC)is a newly emerging paradigm for video compression,which has become a hot research topic home and abroad.Unlike conventional video coding scheme,DVC is based on the distributed source coding of lossless Slepian-Wolf theorem and the lossy Wyner-Ziv theorem.DVC is compliant with the principle of independent encoding and joint decoding,namely DVC explores the statistical property of information source at the decoder.DVC is characterized with simple encoder,complexity decoder and excellent error resilience.DVC is especially applicable to wireless video terminal devices which are usually limited in power and computing.Coding mode decision module is the starting module of DVC,and its performance restricts the overall performance of the system.Based on deep analysis of the influence of the group of pictures(GOP)structure on the coding performance of DVC system,as well as the fixed and adaptive GOP structure algorithms existed,several corresponding solutions have been proposed aiming at imperfection of these algorithms.The main research work of this paper are as follows:1.Several typical DVC frameworks such as pixel-domain Wyner-Ziv system,transform-domain Wyner-Ziv system,DISCOVER,PRISM,and so on,are analyzed.The latest research progresses of mode decision are reviewed.Elaborate research on the influence of the GOP size in fixed GOP structure was conducted.2.The necessity of adaptive GOP structure is analyzed,and a Temporal Correlation Based GOP Size Control Algorithm(TCBG)is proposed.We used two simple and effective indicators,Histogram of Difference(HD)and Block Variance Difference(BVD)to characterize the inter frame correlation and make a quick decision on the GOP structure according to the clustering of the two indicators.Experimental results show that,the TCBG algorithm can effectively improve the performance of the DVC system.3.A Temporal-spatial Correlation Based GOP Size Control Algorithm(TSBG)is proposed.Based on indicators HDand BVD,we defineto represent the temporal correlation of video frames and SIM_t~S to represent the spatial correlation of video frames.Centered on the comparison of the value ofand SIM_t~S,we determine GOP size in two steps-the first step finds a coarse GOP size by progressively looking for a frame which has higher probability of being intra-coded,then,the second step refines it based on iterative checking algorithm to decide the final GOP size,which makes the mode decision more accuracy and effective.Simulation results confirm that the proposed GOP size control algorithm can widely adapte to video sequences with various motion,and at the same time,the proposed algorithm has better RD(Rate Distortion,RD)performance than previous DVC architectures with low system latency and hardly add to the complexity of the system.