El-Shekheby shereen / Port Said University Faculty of Engineering
F. Abdel-Kader Rehab / Port Said University Faculty of Engineering
W. Zaki Fayez / Mansoura University, Faculty of Engineering.
Restoration of spatially-varying blurred images is a challenging task needed in various computer vision systems and applications. In this paper, we propose a novel spatially varying blur detection and restoration method. Motion blur is detected automatically from a single image. Initially, the blur kernel length and direction are estimated by finding the kernel that maximizes the likelihood of a blurred local window. This is achieved by incorporating either vertical or positive diagonal kernels with various lengths. Then, initial blur regions are estimated using a kernel specific feature. Next, the initial blur regions are refined with the support of the image segmentation (CCP) method and neighboring information. Finally, Blurred regions are recovered using the best-estimated kernel. Comparisons with state of the art methods reported in the literature demonstrate accuracy improvements in the image blur detection and restoration results.