An effective total variation image deblurring approach based on enhanced sharp edge prediction

In this paper, a blind deconvolution method which deblurs a single image to produce a salient clear image is proposed. The purpose of our work is to reduce ringing artifacts by using enhanced sharp edges from a recovered latent image, and auto-thresholding to truncate errors of an estimated kernel. Moreover, our method also accelerates conventional iterative deblurring processes. To extract enhanced sharp edges, we first apply some image preprocessing techniques and then an edge mapping method is used to remain accurate edges and smooth regions, which will be both applied to the later kernel estimation. We formulate the optimization function by introducing penalty techniques with image derivatives for kernel and latent image estimation. As a result, experimental results show that our approach is more efficient than conventional methods, and is capable of effectively for producing sharp images with slight ringing effects.