Secret of blind image deblurring?

Blind deblurring from single image is a very hot research topic in the filed of low-level vision with significant applications in CCTV surveillance for public safety, ADAS for intelligent transportantion, and so on. Since the influential work of Fergus et al. (2006, CVPR), blind deblurring research has undergone a rapid development and Korean and Chinese researchers have particularly made great contributions to the field. A comparative study by Lai et al. (2016, CVPR) presents a list of outstanding blind deblurring papers published in ICCV, CVPR and ECCV from 2006 to 2014. The table below (Lai et al., 2016, CVPR) shows the references and the authors. Besides those methods, Pan et al. present three new blind deblurring methods in CVPR, 2016, who are really one of the activest groups on the blind deblurring topic. Actually, Pan-14 (CVPR, 2014; PAMI, 2016) in the table and dark channel prior-based blind deblurring method (CVPR, 2016; also shown in the table) are the two most leading methods among the literature, both of which utilize the l0-norm of image gradients plus addtional critical image priors for reliable and effective blur kernel estimation. 

Now a question arises that what on earth a good image prior is for blind deblurring? Though Pan-14 and Pan-16 as well as others provide answers to the question, I cannot be satisfied with them since those priors are not physically intuitive and simple. What I mean is that those are not blind deblurring image priors in essence. He et al. (CVPR, 2009) proposed the dark channel prior for defogging and earned the best paper award of that year. To me, the dark channel prior is indeed a defogging image prior. And our vision for blind image deblurring is the pursuit of a simple, rational, and workable model in a similar sprite to the dark channel prior. In this perspective, there is a vast space for blind image deblurring research and its practical applications (real-world blurred images with saturation regions and (non-) Gaussian noise).   

  

AlgorithmReferenceFergus-06Fergus et al., Removing camera shake from a single photograph. SIGGRAPH, 2006.Cho-09Cho and Lee. Fast motion deblurring. SIGGRAPH Asia, 2009.Xu-10Xu and Jia. Two-phase kernel estimation for robust motion deblurring. ECCV, 2010.Krishnan-11Krishnan et al. Blind deconvolution using a normalized sparsity measure. CVPR, 2011.Levin-11Levin et al. Efficient marginal likelihood optimization in blind deconvolution. CVPR, 2011.Whyte-12Whyte et al. Non-uniform deblurring for shaken images. IJCV, 2012.Sun-13Sun et al. Edge-based blur kernel estimation using patch priors. ICCP, 2013.Xu-13Xu et al. Unnatural L0 sparse representation for natural image deblurring. CVPR, 2013.Zhang-13Zhang et al. Multi-image blind deblurring using a coupled adaptive sparse prior. CVPR, 2013.Zhong-13Zhong et al. Handling noise in single image deblurring using directional filters. CVPR, 2013.Michaeli-14Michaeli and Irani. Blind deblurring using internal patch recurrence. ECCV. 2014.Pan-14Pan et al. Deblurring text images via l0-regularized intensity and gradient prior. CVPR, 2014.Perrone-14Perrone and Favaro. Total variation blind deconvolution: The devil is in the details. CVPR, 2014.

Pan-16       Pan et al. Blind image deblurring using dark channel prior. CVPR, 2016.