Blur Estimation in Iterative Super-Resolution Restoration Algorithms

It is imperative for imaging model in super-resolution image restoration algorithms to be a good representative of the reality, so that the possible degradations can be corrected. Blur is an important and common degradation. Accurate estimation of the amount of blur has significant effect on the results. Most researchers treat blur as a separate function and do not handle it in the actual restoration algorithm. Iterative algorithms like POCS (projection onto convex sets) and IBP (iterative back projection) use pre-estimated blur parameters. In this work, the blur function is assumed to be Gaussian optic blur and estimation of blur variance is embedded in the iterative algorithm. The histogram distribution of small image blocks is used as a update/correction measure for the variance. In the experimental runs on test images, the blur variance is accurately estimated, through which clear improvements in high resolution images, compared to bicubic interpolation, are obtained