Variational Low-Rank Matrix Factorization with Multi-Patch Collaborative Learning for Hyperspectral Imagery Mixed Denoising

In this study, multi-patch collaborative learning is introduced into variational low-rank matrix factorization to suppress mixed noise in hyperspectral images (HSIs). Firstly, based on the spatial consistency and nonlocal self-similarities, HSI partitioned overlapping patches with a full band. The similarity metric fusing features exploited select most similar construct corresponding patches. Secondly, considering that latent clean holds property across spectra, whereas component does not, proposed Bayesian framework for each patch. Using Gaussian distribution adaptively adjusted by gamma distribution, noise-free data can be learned exploring properties of spatial/spectral domain. Additionally, Dirichlet process mixture model utilized approximate statistical characteristics noises, which constructed exploiting inverse Wishart process. Finally, inference estimate all variables solve using closed-form equations. Widely used datasets different settings are adopted conduct experiments. quantitative qualitative results indicate effectiveness superiority method reducing noises HSIs.