Graph Spatio-Spectral Total Variation Model for Hyperspectral Image Denoising

Abstract

The spatio-spectral total variation (SSTV) model has been widely used as an\neffective regularization of hyperspectral images (HSI) for various applications\nsuch as mixed noise removal. However, since SSTV computes local spatial\ndifferences uniformly, it is difficult to remove noise while preserving complex\nspatial structures with fine edges and textures, especially in situations of\nhigh noise intensity. To solve this problem, we propose a new TV-type\nregularization called Graph-SSTV (GSSTV), which generates a graph explicitly\nreflecting the spatial structure of the target HSI from noisy HSIs and\nincorporates a weighted spatial difference operator designed based on this\ngraph. Furthermore, we formulate the mixed noise removal problem as a convex\noptimization problem involving GSSTV and develop an efficient algorithm based\non the primal-dual splitting method to solve this problem. Finally, we\ndemonstrate the effectiveness of GSSTV compared with existing HSI\nregularization models through experiments on mixed noise removal. The source\ncode will be available at https://www.mdi.c.titech.ac.jp/publications/gsstv.\n