Measurement Matrix And Reconstruction Algorithm Research Based On Correlation Of Pixels In Compressed Image/Video Sensing

The traditional image/video signal coding scheme is based on Nyquist's theorem.Firstly,it is collected with a sampling frequency that is greater than or equal to twice the maximum frequency of the signal.Then,a complex compression coding algorithm is used to compress the collected signal,and a large amount of redundant data is discarded.In order to achieve efficient storage and transmission.This kind of coding framework for performing complex calculations at the encoder side brings huge computing pressure,which is not applicable under the resource constrained application scenario of the acquisition end,such as wireless video surveillance,wireless sensor networks.The Compressed Sensing(CS)theory is an entirely new signal acquisition and encoding/decoding theory that makes full use of signal sparsity or compressibility.It breaks through the limitation of the sampling frequency in the traditional sampling Nyquist theorem,so it has been received extensive attention of the academic community.The research of image/video compressive sensing includes three key technologies: measurement matrix,sparse representation and reconstruction algorithm.Among them,designing measurement and corresponding algorithms based on the features of image/video signals to achieve higher reconstruction efficiency is the focus of image/video compressive sensing research.In this paper,based on the pixel correlation of image and video signals,the compressive sensing measurement matrix of image/video signal and its corresponding reconstruction algorithm are studied in depth.The specific research work includes the following two parts:1.Based on the spatial correlation of image/video,this paper proposes an Adjacent Pixels Correlation Matrix(APM)generation algorithm based on the neighborhood pixel correlation.In this part,two target selection schemes are proposed,and the appropriate kernel functions are used to assign weights,and an image/video measurement matrix with good reconstruction performance is constructed.Simulation experiments show that the proposed measurement matrix has higher image/video reconstruction quality,lower reconstruction complexity,and wider application range.2.The 2s-MHR algorithm includes the two phases of BCS-SPL independent reconstruction and intra-frame multiple hypothesis prediction when reconstructing video keyframes.In the measurement using the APM matrix,the initial estimation of the reconstructed frame and the Wiener filtering in the iterative process have a certain influence on the complexity and reconstruction performance of the BCS-SPL reconstruction algorithm.In order to further improve the video reconstruction performance based on APM matrix measured,this paper combines the initial estimation with Wiener filtering at non-target points on the basis of the reconstruction framework of 2s-MHR,and proposes an improved framework of SPL.Simulation experiments show that the framework has better reconfiguration performance under the precondition of APM matrix measured.