Video Compressed Sensing Based On Surfacelet

With the development of wireless networks, video applications in wirelesscommunication become increasingly important, but traditional methods can not meet therequirements of low-power and low complexity on the encoding side in the wireless videocommunication. Compressed sensing theory appears to solve the above problems byproviding new ideas, merging video sampling and compressing together, it reducs thecomputational complexity of encoding side effectively.Current video compressed sensing algorithms are divided into two categories: framereconstruction and overall reconstruction. With the shortcoming of frame reconstructionby which there exits inter-frame jitter effect in the reconstructed video taken intoconsideration, video was regarded as a three-dimensional signal and was reconstructed byoverall reconstruction in my paper. Surfacelet transform was used as sparse representationof a video, combined with the non-local similarity to suppress noise and retain videoimage details better, then a video compressed sensing algorithm based on Surfacelet andnon-local similarity was proposed and experiments were done to demonstrate theeffectiveness of the algorithmIn the existing video compressed sensing, the lack of sparse representation flexibilityof video by pre-configured basis functions affects the reconstruction quality. To addressthis issue, a video compressed sensing algorithm based on classified dictionary ofSurfacelet was proposed. Surfacelet was used to a video sequence to get three-dimensionalsparse decomposition, and the resulting high frequency coefficients were divided intosub-blocks, by clustering blocks of similar structure and PCA algorithm classifieddictionary of Surfacelet was obtained. During the reconstruction of video, the dictionarywas used as sparse representation, in order to further accelerate the convergence speed ofthe algorithm, variable splitting was used to divided one spatial variable into spatial andfrequency dual variables, and through augmented Lagrange method approaching theoptimal solution alternately to reconstruct video. Experimental results show that thealgorithm can reconstruct the original details of the video effectively, and the reconstruction of video moving intensely is better than other algorithms.