A novel independent deconvolution method for blurred and partially saturated images

Motion deblurring is a knotty problem, especially when the photo is shot in a low lighting scene with a long exposure time or a strong reflected lighting scene. In this situation, partially saturated pixels violate the assumption of linear model and the recovered image contains severe ringing artifacts. In this paper, we proposed a novel two-layer images independent deconvolution method that redefine the shift-invariant motion blur model and assume the blurry image composed of unsaturated foreground and saturated background to handle partially saturated images. In order to reduce ringing artifacts, we expanding the unsaturated region to saturated. For saturated background, we make the saturated region shrinkage according to the modified Richardson-Lucy algorithm. Then, these two layers use two different deconvolution algorithms respectively to get a high quality latent image. Experimental results on synthesized and real-life images showed that this approach is competitive with other state-of-the-art algorithms.