Image Registration with Linear Distribution-Based Adaptive Differential Evolution Algorithm

Image registration can be viewed as an optimization problem based on certain similarity measurement. Accuracy in image registration is still an open issue. In order to seek global optimality, population-based differential evolution algorithm is a very effective algorithm. How to balance the exploration and exploitation abilities of algorithms is one of the important factors affecting algorithm accuracy. The control parameter CR plays a balancing role in the adaptive differential evolution algorithm. For image registration applications, an effective method of adaptive generation of CR value is proposed. When the CR value is large, the algorithm has strong global search ability; On the contrary, when the CR value decreases, the algorithm's local search ability becomes stronger. Therefore, this paper designs three linear functions according to the change of control parameters CR to meet the different development stages of the population. At the beginning stage of the search, the probability density distribution of parameter CR increases linearly. At this time, the global search ability of the algorithm should be strengthened to maintain population diversity. As the iteration progresses, the probability density function of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$C R$</tex> is adjusted to a uniform distribution. Then, the population keeps updating, and PDF will become a linear decreasing function. This method compares the accuracy of image registration with other 10 advanced DE algorithms and metaheuristic optimization algorithms based on population development in medical images and natural images. The results show that the algorithm has the best performance in terms of search accuracy compared with other algorithms.