A method for blind image deblurring based on Gabor filter edge estimation

Edge is widely used in blind deconvolution methods to provide the information source for the deblurring process. In the natural images, the blurring can be shown in any directions on edges randomly. However, the previous edge-based deconvolution methods made use of vertical and/or horizontal edge information only in the recovering process that lowers the performance. In this paper, a novel method of blind deconvolution using Gabor filter to estimate the edge information in omnidirectional directions is proposed. The deblurring process is then followed by applying fast iterative shrinkage-thresholding algorithm and iteratively reweighted least squares to recover the true sharp image. Also, the paper proposes a sharpness measurement, named as Haar defocus score, based on Haar-wavelet transformation, to estimate the quality of the deblurred image in cases no ground-truth exists for comparison. Experiments on the common public databases in the field show promising performance of the proposed method with respect to both PSNR and Haar defocus score measurements in comparison with the previous methods.