文章基本信息

标题：Blind Image Deblurring via Regularization and Split Bregman
本地全文：下载
作者：Blind Image Deblurring via Regularization ; Split Bregman
期刊名称：IAENG International Journal of Computer Science
印刷版ISSN：1819-656X
电子版ISSN：1819-9224
出版年度：2018
卷号：45
期号：4
页码：584-591
出版社：IAENG - International Association of Engineers
摘要：Degradation caused by blurring is ubiquitous indigital images, and blind image deblurring (BID) has been proposedto solve this issue. Over the past several decades, varioustechniques and tools have been developed for BID problems,and continual efforts have been made to improve the speed andaccuracy of sharp image estimation. This study proposes a newBID method, which incorporates a regularization technique,sparsity-inducing priors, and the split Bregman method. In thefirst phase, the proposed method equates blur estimation witha constrained minimization problem in which sparsity-inducingpriors are employed to regularize the gradient image and blurthe kernel. The split Bregman method is then applied to divideand conquer the minimization problem to optimize the blur estimation.To enhance the accuracy of the outputs, a coarse-to-fineupdating procedure is integrated into the Bregman iterations.The resulting subproblems are efficiently addressed during thealternating iteration by employing methods such as the fastFourier transform (FFT) and hard shrinkage. In the secondphase, the total variation (TV) deconvolution model is appliedto sharp image reconstruction, and a classic half-quadraticapproach is applied to handle the model with high efficiency.In our experiments, the proposed method and three similarmethods are employed to deal with synthetic blurry imagesand real-world blurry images from open image databases. Thedeblurring results are presented in the form of recovered imagesand peak signal-to-noise ratio (PSNR) values. To compare speedperformances, the computation times for image deblurring arecomputed and reported. The proposed method can be applied toefficiently handle various types of blurry images and producesatisfactory outputs. Experimental outputs indicated that theproposed method provides superior restoration quality andcomputing speed compared with alternatives.
关键词：blind deblurring; regularization technology;sparse inducing; split Bregman; hard shrinkage