Abstract: As the two-stage multi-hypothesis reconstruction ( 2sMHR) for the prediction in both measurement and pixel domains is of low reconstruction quality at low subrate as well as low prediction accuracy and high computa- tional complexity for fast movement sequences,on the basis of 2sMHR scheme,a novel iterative multi-hypothesis ( IMH) prediction reconstruction algorithm utilizing fast searching technology is proposed.This algorithm utilizes the flexibility of block partition in pixel domain to conduct overlapping block-based iterative multi-hypothesis pre- diction,and adopts the scheme of jointing cross-shaped and regional searching to lessen matching complexity in large-range searching for fast sequences.In addition,by utilizing the space correlation in images,an adaptive in- terframe/intraframe hypothesis-block selecting scheme is presented.In comparison with the existing algorithms,the proposed IMH prediction reconstruction algorithm suppresses the prediction block effect,reduces the computational complexity for large-range searching and improves the video reconstruction quality.

Key words: compressed video sensing, multi-hypothesis prediction, hypothesis block, pixel domain, iteration prediction, prediction accuracy, time complexity