A Novel Subspace Super-Pixel Based Low Rank Representation Method for Hyperspectral Denoising

摘要

This paper presents a novel denoising method based on subspace superpixel based low rank representation for hyperspectral imagery. First, the original hyperspectral data is assumed to be low-rank in both spectral and spatial domains. The spectral low rankness of HSI data is represented by decomposing it into two sub-matrices of lower rank while the spatial low rankness is explored within superpixel based regions in the subspace. The superpixels are generated by utilizing state-of-the-art superpixel segmentation algorithms in the first principle component of the original HSI. The final model could be efficiently solved by augmented Lagrangian method (ALM). Experimental results on simulated hyperspectral dataset validate that the proposed method produces superior performance than other state-of-the-art denoising methods in terms of quantitative assessment and visual quality.