Estimation of the noise variance of uniform linear motion blurred images

The key problem in restoration of motion blurred images is how to get the point spread function and the noise information. It is commonly assumed as the white Gaussian noise. The paper proposes a method for estimating the variance of noise in uniform linear motion blurred images. A superposition operator on the difference of the motion-blurred image is used to produce a noise-dominated image. This method can keep noise in different columns of the noise-dominated image irrelevant. There lies numerical relationship between the variance of superposition and original noise. The variance of original noise can be computed by using the numerical relationship. Experiments prove that the noise power can be computed accurately by utilizing this method.