Unified regularization framework for blind image super-resolution

Blind superresolution (BSR) is one of the challenges in image superresolution. We propose a new approach using a unified regularization framework, which solves image registration, point spread function (PSF) estimation, and high-resolution (HR) image reconstruction simultaneously. To achieve this, the anisotropic diffusion techniques are employed as one regularization term to preserve edge information in the HR image estimation, and a generalized version of the eigenvector-based (EVAM) constraint is developed to regularize the PSF. An alternating minimization algorithm is devised to find optimal solutions, and an effective numerical implementation scheme, based on local filtering, is proposed to suppress the ringing artifacts in the image reconstruction. Finally, experiments with synthetic and real data are presented to demonstrate the effectiveness and robustness of our approach, which can handle motion blur well and enhance resolution notably for very noisy images. (C) 2007 Society of Photo-Optical Instrumentation Engineers.