Blur kernel estimation using blurry structure

Abstract

Motion deblurring has been a hot spot of research given its wider application range. It has been proven to be effective in several recent studies that utilize structure in the intermediate image to estimate a blur kernel. However, these methods ignore to extract blurry structure from input blurry image. This will cause imbalance in the objective function and introduce artifact errors into the deconvolution process. In this paper we will first exploit a mask determined by convolution of an intermediate image with a kernel to generate blurry structure, and then take it into data term instead of blurry image to overcome the problem. Moreover, we employ sparse prior of kernel and propose a novel L0 regularization for accurate kernel estimation. Experiments across datasets showed that our algorithm achieved the state-of-the-art motion deblurring results.