The Optimization of Extended Hypothesis Set for Compressive Video Sensing

Abstract

Almost all existing multihypothesis (MH) prediction methods in compressive video sensing (CVS) are absorbed in exploiting the reference information in key frames to guide the reconstruction of non-key frames. However, when the non-key frames are distant from key frames, the temporal correlation between them declines. To address this problem, we consider the non-key frames reconstructed before the current frame and extend the hypothesis set in MH prediction by the hypotheses extracted from them. Then, to avoid the high computational complexity caused by the large size of extended hypothesis set, a novel optimization technique for hypothesis set in MH prediction is proposed. Experimental results show that the strategy we proposed outperforms the state-of-the-art technique in reconstruction quality within an acceptable computational complexity.