Research On Snapshot Compressive Imaging Algorithm Based On Sparse Dictionary Learning

The fact that many existing optical sensors can only acquire bandwidth-limited two-dimensional spatial information hinders machines from effectively perceiving high-dimensional data into the real world.Snapshot compressive imaging(SCI)can effectively solve the above problems,that is,the optical imaging system is used to carry out two-dimensional compression measurements of high-dimensional data,and the sparseness of image is used to reconstruct high-dimensional data through corresponding compressive sensing algorithms.Plug-and-play(PnP)framework is a novel approach to solve the problem of SCI reconstruction,but existing denoising process still faces many challenges,such as requiring matching data.Aiming at these image reconstruction problems of SCI,this paper improves PnP algorithm and proposes a new algorithm framework: PnP algorithm based on K-means singular value decomposition(KSVD).Compared with generalized alternating projection algorithm based on total variation,the peak signal-to-noise ratio value is increased by 2d B on average and structural similarity is increased by 0.02 on average in solving problems of video reconstruction.When solving hyper spectral imaging problems,the peak signal-to-noise ratio value is increased by 3d B.