HYPSPECTRAL IMAGE DENOISING VIA MULTIDIMENSIONAL NONLOCAL MODEL

Hyperspectral images (HSIs) often suffer from various noise. Noise reduction is a crucial task for improving HSIs quality. Recently, the nonlocal means (NLM) shows the good performance for preserving detail information and removing noise. In this paper, we propose a hyperspectral image denoising algorithm taking advantage of the redundancy from both spectral and spatial domain We extend 2D nonlocal denoising model to multidimensional nonlocal model. The model takes account of the two important forms, including multidimensional filter model and variational model. In filter model, the similar image patches are searched in HSI cube because of the high correlation between bands. In the variational model, this paper establishes a maximum a posterior (MAP) framework for HSI denoising by introducing a multidimensional nonlocal total variation (MNLTV) prior. The proposed prior takes full advantage of the redundancy and continuity of HSI. Experiments with synthetic hyperspectral datas illustrate that the proposed method can obtain better denoising results for hyperspectral datas than traditional nonlocal approach.