Noise Adaptive Image Restoration: A Minimum Mean Square Error Approach

This paper introduces a novel minimum mean square restoration approach with noise adaptive regularization for linear, space invariant known blur. The restoration filter is designed to de-blur the image while minimizing the noise in the image. The adaptive noise reducing characteristic of the filter is proved analytically. It has been substantiated with empirical study that this restoration approach outperforms state of the art restoration approaches. The proposed method has two major advantage: i) It de-noises the image based on the characteristic of the individual image, ii) It has simple, non-iterative implementation.