Research On Compressed Image/Video Sensing Reconstruction Algorithm Based On Structural Feature

Based on the Nyquist sampling theory,the video coding technology needs to first sample a signal at a rate higher than twice the highest frequency of the signal,and then remove the redundant information through compression coding with high computational complexity.The operation process of highly compressing following highly sampling is not only a huge waste of resources at the encoder,but also exerts a huge computational pressure at the decoder,which is no longer applicable to the resource-limited application scenario.Compressed sensing theory combines the sampling and compression process,greatly simplifying the coding process and transfering the computational pressure from encoder to decoder.A video signal consists of multiple image signals with correlation.This paper focuses on the research of compressed image/video sensing reconstruction algorithm.Most video reconstruction algorithms reconstruct some video frames by using image reconstruction algorithm and reconstruct the others by using inter-frame correlation.Most of the more prominent image/video reconstruction algorithms involve similar blocks matching,and the final reconstruction quality of the image/video signal is greatly affected by the similar block matching degree.In this paper,for the case of poor condition in similar block matching,the corresponding solutions are proposed to reconstruct the image and video signals by uutilizing the structural features of the image blocks.The specific research work is divided into the following two parts:1.At low sampling rates,image reconstuction algorithms suffer initial reconstruction of poor quality,leading to inappropriate grouping situations and an unsatisfying reconstruction result.Based on GSR,a low-rank enhancement reconstruction algorithm named LRER is proposed in this paper.A hybrid filtering reconstruction method is proposed to improve initial reconstruction and then a low-rank enhancement pretreatment is conducted before similar-block grouping to help grouping operation focus more on the key primary features,which improves the rationality of the grouping result.The simulation results show that the proposed algorithm significantly improves reconstruction performance at low sampling rates.2.In multi-hypothesis prediction-based compressed video sensing reconstruction algorithms,the matching degrees of the hypothesis set corresponding to different image blocks are quite different,so the reconstruction difficulty of different blocks is obviously different.In this paper,a multi-hypothesis-based local enhancement reconstruction algorithm named MHLE is proposed.Image blocks are classifying into two categories and a pixel domain dual channel matching strategy is proposed for moving image blocks,where the basic features of the image blocks are enhanced to improve the matching effectivity and obtain a higher quality hypothesis set.Besides,the structural similarity evaluation criteria is introduced into the matching block weight assignment process to improve prediction accuracy.The simulation results show that the reconstruction quality of the proposed algorithm is superior than other multi-hypothesis prediction-based reconstruction algorithms.Compared with the group sparsity-based reconstruction algorithms,the proposed algorithm possesses faster reconstruction speed and higher reconstruction quality at most sampling rates.