Hyperspectral Image Denoising Based On Non-convex Local Low-rank Sparse Separation And Spatial-spectral Total Variation Regularizatio

Hyperspectral images have rich spatial structure information and are widely used in many fields.Because of sensor internal malfunction,photon effects,atmospheric interference and many other factors,hyperspectral images are often contaminated by various types of noise,such as Gaussian noise,deadline,salt and mixed noise.The existence of hybrid noise in hyperspectral images severely degrades the data quality,the interpretation accuracy of hyperspectral images,and the subsequent hyperspectral image applications.Therefore,there is an urgent need to denoise hyperspectral images.The local low-rank properties of hyperspectral images,low-rank techniques were proposed and successfully applied to hyperspectral image denoising.However,traditional low-rank techniques do not fully consider the spatial structure information and spectral properties of hyperspectral images.Therefore,based on the low-rank and spatial-spectral properties of hyperspectral images,this thesis proposes a new denoising model with the logarithmic approximation-based non-convex RPCA and spatial spectral total variation regularization（L³S³TV）.The research content of this thesis is as follows:（1）We propose a novel non-convex approach to robust principal component analysis for hyperspectral image denoising,which focuses on simultaneously developing more accurate approximations to both rank and column-wise sparsity for the low-rank and sparse components,respectively.（2）The new method adopts log-determinant rank approximation and a novell_2,lognorm,to restrict the local low-rank or columnwisely sparse properties for component matrices,respectively.For thel_2,log-regularized shrinkage problem,we raise an efficient,closed-form solution,which is namedl_2,log-shrinkage operator.The new regularization and the corresponding operator can be generally used in other problems that require column-wise sparsity.（3）Moreover,we impose the spatial-spectral total variation regularization in the log-based non-convex RPCA model,which enhances the global piece-wise smoothness and spectral consistency from the spatial and spectral views in the recovered hyperspectral image denoising.（4）Extensive experiments on both simulated and real data sets demonstrate the effectiveness of the proposed method in hyperspectral image denoising.