Single-image motion deblurring using adaptive anisotropic regularization

摘要

Abstract. We present a novel algorithm to remove motion blur from ansingle blurred image. To estimate the unknown motion blur kernel asnaccurately as possible, we propose an adaptive algorithm using aniso-ntropic regularization. The proposed algorithm preserves the point spreadnfunction u0001PSFu0002 path while keeping the properties of the motion PSF whennsolving for the blur kernel. Adaptive anisotropic regularization and refine-nment of the blur kernels are incorporated into an iterative process tonimprove the precision of the blur kernel. Maximum likelihood u0001MLu0002 esti-nmation deblurring based on edge-preserving regularization is derived tonreduce artifacts while avoiding oversmoothing of the details. By using thenestimated blur kernel and the proposed ML estimation deblurring, thenmotion blur can be removed effectively. The experimental results for realnmotion blurred images show that the proposed algorithm can removesnmotion blur effectively for a variety of real scenes. © 2010 Society of Photo-nOptical Instrumentation Engineers. u0003DOI: 10.1117/1.3487743