Correlation-based Iterative Estimation of Distortion Kernel for Image Restoration

Abstract

The main goal of the work is to develop an algorithm for automated image restoration. The method of analyzing image quality is based on the analysis of line-to-line and column-to-column correlation coefficients. Using such quality measures the refinement of the distortion operator is performed iteratively based on the Lucy-Richardson method for image reconstruction. The estimate of the distorting operator is found using gradient descent. The experiments showed that applying the proposed method to the images having smear and blur distortion gives correct restoration results. Since the approach uses optimization methods of directional search, it is also possible to reduce the number of necessary calculations.