Map-Based Denoising of Hyperspectral Imagery Using 3-D Edge-Preserving Priors

In the hyperspectral imaging, acquired images are inherently affected by noise, whose levels may vary from band to band. It is not a trivial task to remove this kind of noise while preserving the edges and details of hyperspectral images (HSIs). This paper provides a maximum a posterior (MAP)-based denoising approach for HSIs corrupted by band-varying noise. Compared with the classical MAP-based methods for 2-D degraded image restoration, the proposed approach uses 3-D edge preserving priors to keep sharp edges while smoothing the 3-D HSIs. In order to adapt to the characteristics of bandvarying noise statistics and high dynamic ranges of HSIs, we adaptively estimate the noise variance and scaling parameter of each point. For minimizing the cost function, the half-quadratic optimization algorithm is used. Both denoising and classification experimental results confirm the superiority and validity of the proposed method.