Research On Group-based Sparse Reconstruction Algorithm For Compressed Video Sensing

Traditional video coding technology uses Nyquist's theorem to sample,that is,the sampling frequency should be more than or equal to twice the maximum frequency of the signal,and then complete the compression process by removing the redundancy of a large number of sampled signals.This video compressed coding method requires strong computational power and will cause waste of resources at the sampling end.It is not suitable for the situation where the resources are limited at the sampling end.Compressed Sensing theory combines sampling and compression,breaks through the limitation of Nyquist theorem on sampling frequency,completes compression while sampling,and transfers computational complexity from sampling end to reconstruction end.Compressed Video Sensing is the focus of compressed sensing theory research in recent years.It is a new video acquisition method based on Compressed Sensing theory.It is especially suitable for situations where the resources of sampling end are constrained and the computing power is limited.1.By researching on the SSIM-InterF-GSR algorithm,an adaptive threshold algorithm based on sampling rate(AT-GSR)is proposed in group-based sparse representation.The simulation results show that the proposed reconstruction algorithm AT-GSR reduces the complexity of the algorithm compared with SSIM-InterF-GSR and improves the reconstruction performance for slow moving video sequences.Compared with the current 2sMHR and RRS algorithms,the reconstruction performance of AT-GSR is also significantly improved.2.In view of the limitation of AT-GSR threshold setting,a classification threshold processing algorithm based on mean square error of measurments(yMSE-CTP-GSR)is proposed.The video sequence is adaptively set a suitable threshold according to the initial prediction performance,and the threshold decreasing scheme is selected.An adaptive reference frame scheme based on frame motion is proposed.The simulation results show that yMSECTP-GSR algorithm has good reconstruction effect for all of the fast motion sequences,slow motion sequences and motion variation sequences,with an average increase of 1-2 dB.3.Aiming at the poor reconstruction effect of AT-GSR algorithm at low sampling rate,an intra-group optimal block selection and recovery scheme(OB-S/R)based on structural similarity is proposed.This scheme uses the characteristics of structural similarity to improve the block similarity in the group again to increase the group sparsity.At the same time,the block with reduced similarity is discarded and the reconstructed frame accuracy is increased.The simulation results show that the reconstruction effect of AT-GSR algorithm is greatly improved at low sampling rate,with an average increase of 1-3 dB.