Interleaved Regression Tree Field Cascades for Blind Image Deconvolution

Image blur from camera shake is a common cause for poor image quality in digital photography, prompting a sig-nificant recent interest in image deblurring. The vast ma-jority of work on blind deblurring splits the problem into two subsequent steps: First, the blur process (i.e., blur ker-nel) is estimated; then the image is restored given the esti-mated kernel using a non-blind deblurring algorithm. Re-cent work in non-blind deblurring has shown that discrim-inative approaches can have clear image quality and run-time benefits over typical generative formulations. In this paper, we propose a cascade for blind deblurring that alter-nates between kernel estimation and discriminative deblur-ring using regression tree fields (RTFs). We further con-tribute a new dataset of realistic image blur kernels from human camera shake, which we use to train the discrim-inative component. Extensive qualitative and quantitative experiments show a clear gain in image quality by inter-leaving kernel estimation and discriminative deblurring in an iterative cascade. 1