Deblurring Text Images Using Kernel Dictionaries

Image deblurring is one of the widely studied and challenging problems in image recovery. We propose a new approach to represent and compute a blurring kernel, and provide a novel method to recover text images affected by multiple kernels. We utilize some unique kernel structures to attack problems where these kernels can be estimated using linearly weighted combinations. By exploiting these structures, we provide fast and accurate solutions when dealing with certain kernel types. We demonstrate the usefulness of this methodology through computational study.