A deep variational Bayesian framework for blind image deblurring

Blind image deblurring is an important but challenging problem in processing. Traditional optimization-based methods typically formulate this task as a maximum-a-posteriori estimation or variational inference problem, whose performance highly relies on handcrafted priors for both the latent and blur kernel. In contrast, recent deep learning generally learn from large collection of training images. Deep neural networks (DNNs) directly map blurry to clean kernel, paying less attention physical degradation process image. study, we present Bayesian framework blind deblurring. Under framework, posterior kernel can be jointly estimated amortized manner with DNNs, involved DNNs trained by fully considering model, supervision data driven which naturally led lower bound objective. Comprehensive experiments were conducted substantiate effectiveness proposed framework. The results show that it achieve promising relatively simple incorporate existing enhance their performance.