Multi-scale Deblurring with Smooth Region Constraints

In this paper, a new multi-scale deblurring method is proposed to remove the motion blur. The method estimates the blur kernel by an alternative algorithm at scales from coarse to fine. After the blur kernel is estimated in the finest scale, the blurred image is restored via image deconvolution. To remove the ringing artifacts, we propose a smooth regions constraint. Combining with the noise prior, image statistical prior and image sparse representation, the smooth region constraints can be used to suppress the ringings in the image deconvolution. The proposed method can estimate the accurate blur kernel and restore the high quality image with ringing removing. Various comparative results demonstrate the efficiency of our method.