Blind image deconvolution via salient edge selection and mean curvature regularization

Blind image deconvolution is an ill-posed problem since there exists infinite pairs of blur kernels and latent images. To obtain reasonable results this problem, most previous methods have emphasized the importance selecting salient edges for kernel estimation. In paper, a blind method based on explicit implicit selection proposed. Explicit edge achieved by using mutually guided filtering, while mean curvature regularization adopted to remove disadvantageous structures implicitly. Besides, proposed model can be efficiently optimized half-quadratic penalty filter. Extensive experiments show that performs favorably against state-of-the-art both synthetic real-world blurred