Ensemble Learning and Bi-splitting Bregman Regularization for Image Deblurring

In the process of imaging,camera shake and defocusing during exposure are extremely common reasons for image degradation,and lead to objectionable image blur.Due to images of real-world scenes obey heavy-tailed distributions in their gradients,a mixture-of-Gaussians model is built to represent the distribution.Then Ensemble Learning approach based on the variational Bayesian estimation is applied to find the blur kernel.Given the kernel, the image deconvolution is completed using Total Variation Regularization, and bi-splitting Bregman iterative algorithm under Bregman framework is proposed in this paper.The experimental results show that this algorithm can effectively remove image blur compared to existing approach.