Adaptive key-frame selection based on image features in Distributed Video Coding

The algorithm of periodical key-frame selection (PKFS) in Distributed Video Coding ignores the correlation of adjacent frames so as to fail in time redundancies exploring. According to this situation, we propose a new key-frame selection strategy based on image features which takes advantage of gray accumulate histogram (GAH) and edge direction histogram (EDH) to measure video motion and then judge the key frame whose feature difference lays above the threshold, realizing adaptive decision. Additionally, we add an extra restriction on GOP (Group of Picture) size after observing effect from decoder interpolation process. This algorithm has low computation complexity and experiment results from different sequences show that 0.6-2dB rate-distortion (RD) performance improvement has been achieved and the limitation about GOP size also brings 0.2-1dB advancement.