In blind motion deblurring, leading methods today tend towards highlynon-convex approximations of the l0-norm, especially in the imageregularization term. In this paper, we propose a simple, effective and fastapproach for the estimation of the motion blur-kernel, through a bi-l0-l2-normregularization imposed on both the intermediate sharp image and theblur-kernel. Compared with existing methods, the proposed regularization isshown to be more effective and robust, leading to a more accurate motionblur-kernel and a better final restored image. A fast numerical scheme isdeployed for alternatingly computing the sharp image and the blur-kernel, bycoupling the operator splitting and augmented Lagrangian methods. Experimentalresults on both a benchmark image dataset and real-world motion blurred imagesshow that the proposed approach is highly competitive with state-of-the- artmethods in both deblurring effectiveness and computational efficiency.
展开▼
