Feedback Loop Design and Analysis for Iterative Localized Image Deblurring

This paper considers image deblurring algorithm based on model-based iterative update of the image estimate. The update has two principal terms: (i) integral feedback of blurred image prediction error and (ii) feedback of the past deblurred image estimate. The two feedback terms are computed by applyingtwo FIR operators. The paper shows how these feedback update operators can de designed with a given degree of localization to provide an optimal tradeoff between design objectives such as convergence speed, noise amplification, quality of image restoration, and robustness. The solution to the feedback operator design problem is obtained by Linear Programming optimization. A parallelized implementation of the update algorithm using localized feedback operators is scalable to very large image sizes.