Single-image motion deblurring using an adaptive image prior

Abstract

Blind deblurring is the restoration of a sharp image from a blurred image when the blur kernel is unknown. Most image deblurring algorithms impose a uniform sparse gradient prior on the whole image, and reconstruct the image with piecewise smooth characteristics. Although the sparse gradient prior removes ringing and noise artifacts, it inevitably removes mid-frequency structures, leading to poor visual quality. The gradient profile of fractal-like structures is close to a Gaussian distribution, and small gradients from such regions are severely penalized by the sparse gradient prior. In this paper, we introduce an image deblurring algorithm that adapts the image prior to the underlying detailed structures. The statistics of a local detailed structure can be different from those of the global structure. By identifying the correct image prior for each pixel in the image, our approach models the spatially varying motion blur exhibited by camera motion more effectively than conventional methods based on space-invariant blur kernels. Using different priors for the local region and the motion blur kernel, we derive a minimization energy function that alternates between blur kernel estimation and deblurring image restoration until convergence. Experimental results demonstrate that the proposed approach is efficient and effective in reducing motion blur in an arbitrary direction in a single image.