Hyperspectral Image Denoising Based on Superpixel Segmentation Low-Rank Matrix Approximation and Total Variation

In this paper, we proposed a new method to remove mixed noises in hyperspectral images (HSI's) denoising using superpixel segmentation, low-rank matrix approximation and total variation (SSLRTV). According to the spectral correlation of the HSI bands, it has a low-rank structure in spectral-domain. So at first, we divide the HSI to the homogeneous regions by superpixel segmentation to save the spectral signature of pixels in the low-rank approximation method. Furthermore, each segmented region's rank is estimated to determine the principal spectral subspace. We improved algorithm performance by proposing a new TV model for HSIs that saves spatial and spectral smoothness of the HSI; furthermore, it has a fast convergence speed and simple computational based on the gradient descent method. In the proposed SSLRTV method, the optimization problem is solved by an augmented Lagrange multiplier method. Experiments on the real data and simulated data demonstrate that the proposed denoising method has better results than previous in terms of quality and run-time cost.