Blind Deconvolution for Restoring Blurred and Distorted Images Based on the Pyramid Approach and Genetic Algorithms

Abstract

Automatic restoration of blurred and distorted images is one of the most urgent tasks of image processing. There are two main classes of methods for solving this problem – deep learning and optimization methods. Optimization methods are a classic way to solve this problem and are aimed at selecting an unknown distorting function. This task is called blind deconvolution. The advantage of these methods is their accuracy and quality of the result obtained. However, a significant disadvantage of these methods is their operation time, which can reach more than a dozen minutes. In most methods of this class, classical gradient optimization methods and their modifications are used. However, this approach has a number of disadvantages. One of them is that the objective function must be differentiable. Another significant disadvantage of these methods is the probability of stopping at a local extremum and not finding the optimal solution. Thus, the disadvantages of the step-by-step optimization methods used are added to the long operation time of these methods. Optimization models devoid of these disadvantages (except for optimization time) are genetic algorithms. In this paper, we propose a method of blind deconvolution based on a pyramid approach and the use of a genetic algorithm with a modified mutation operator as an optimization model.