Hyperspectral image denoising from an incomplete observation

Hyperspectral image (HSI) contains rich spectral information, which can facilitate lots of vision based tasks related with immersive communications. However, HSI is easily affected by different factors such as noise, missing data, etc., which degrades the image quality of HSI and makes HSI incomplete. In this study, to guarantee the denoising method can be used for incomplete data and suppress multiple kinds of noise, we analyze HSI denoising as a low-rank matrix analysis (LRMA) problem taking advantage of Hyperspectral unmixing, and model LRMA for HSI denoising probabilistically. A Bayesian LRMA method is then introduced to solve the probabilistic LRMA problem. The proposed method can denoise the noisy incomplete HSI more effectively compared with several denoising methods. Experimental results demonstrate the effectiveness of the proposed method.