Robust Kernel Estimation in Blind Deconvolution

Due to the loss of information about image and the interference of noise, blind deconvolution is an ill-posed problem. In this paper, we study this problem based on the algorithm of Krishnan et al.[1], which uses a normalized sparsity measure to solve the problem. By assuming the random high frequency property of the difference between true kernel and intermediate estimated kernel, we add a Gaussian smoothing filtering during sharp image update step. The filtering process can improve robustness of the algorithm. Experimental results show that our algorithm estimates more precise kernel and run fast than Krishnan's original algorithm.