Noise Separation from Multiple Copy Images Using the FastICA Algorithm

Abstract

This paper proposes an effective method to separate noise from multiple copy images (MCIs). Suppose that noise and original image are mutually independent in mixed signals, the mixed signals are thus decomposed to an original image independent component and a noise component by using fast independent component analysis (FastICA). The original image independent component is selected to reconstruct the resulting image according to the standard deviation of its time course. By modeling the noise as Gaussian, experimental results show that zero-mean and nonzero-mean Gaussian noises can be separated effectively from multiple copy images by the proposed method, which is effective in the case of stable and unstable noise intensity.