MULTISPECTRAL IMAGE DENOISING ALGORITHM BASED ON NONLOCAL LOW-RANK AND TOTAL VARIATION
Multispectral image(MSI)is often contaminated with Gaussian noise during imaging,which affects the subsequent applications.To remove Gaussian noise,by considering the global correlation(GCS)along the spectrum and the non-local self-similarity(NSS)across the space,a new denoising method based on tensor is proposed.To capture both nonlocal similarity and spectral correlation,MSI is first segmented into overlapping three-dimensional full-band patches(3D FBPs),and similar patches are grouped by clustering algorithm.Then each 3D FBP is expanded into a matrix,and the similar patches in the group are cascaded into a third-order tensor,which can be regularized by tensor nuclear norm.To avoid the ringing effect caused by this operation,the three-dimensional weighted total variation is used to explore the spectral-spatial smoothness.Simulation experiment results show that the proposed algorithm effectively explores the inherent GCS and NSS knowledge,and recovers more detailed information from the degraded MSI,which is superior to the comparison methods under the comprehensive quantitative performance index.
万方数据
维普
