Y.-S. Ho (Ed.): PSIVT 2011, Part II, LNCS 7088, pp. 240–251, 2011. © Springer-Verlag Berlin Heidelberg 2011 Blind Image Deblurring with Modified Richardson-Lucy Deconvolution for Ringing Artifact Suppression Hao-Liang Yang, Yen-Hao Chiao, Po-Hao Huang, and Shang-Hong Lai Dept. of Computer Science, National Tsing Hua University, Hsinchu, Taiwan lai@cs.nthu.edu.tw Abstract. In this paper, we develop a unified image deblurring framework that consists of both blur kernel estimation and non-blind image deconvolution. For blind kernel estimation, we propose a patch selection procedure and integrate it with a coarse-to-fine kernel estimation algorithm to develop a robust blur kernel estimation algorithm. For the non-blind image deconvolution, we modify the traditional Richardson-Lucy (RL) image restoration algorithm to suppress the notorious ringing artifact in the regions around strong edges. Experimental re- sults on some real blurred images are shown to demonstrate the improved effi- ciency and image restoration by using the proposed algorithm. 1 Introduction Motion blur is caused by relative motion between the camera and the scene during exposure. The real camera motion is usually too complicated to estimate from a blurred image when it involves camera rotation or large scene depth variations. To simplify the problem formulation, previous researches usually assumed the camera motion is perpendicular to the optical axes and the effect of scene depth variation can be neglected. In other words, the blur kernel is assumed to be spatially invariant. Un- der this assumption, a blurred image, B, can be modeled as (1), where K is the blur kernel, I is the clear image, N is the noise, and  is the convolution operator. ܤൌ ܫ ܭ൅  . (1) The blind image restoration problem in (1) is ill-posed because I and K are highly un- der-constrained and there are infinitely many possible combinations of I and K such that their convolution is equal to the blurred image B. Fergus et al. [4] proposed to utilize ensemble learning to estimate the blur kernel with a sophisticated variational Bayes inference algorithm, which employs the property of specific distributions of image gra- dients for natural images to approximate the posterior distribution. Levin [6] also exploited image statistics for estimating blur kernels. Shan et al. [9] proposed two prob- abilistic models to improve image restoration. One is to model the spatially random distribution of noise, and the other is a smoothness prior model which can reduce the ringing artifacts. Cho and Lee [11] proposed a latent image prediction step, which applied shock filter to recover the sharp edge information for estimating the blur kernel.