关键词: Powell算法, 归一化互信息, 遗传算法

Abstract: Aiming at the Powell algorithm dependent on the initial value and easy to fall into the local extremum, genetic algorithm is applied to improve the Powell algorithm in image registration. Using normalized mutual information as fitness, genetic algorithm searches for image registration parameters in parallel and global area. Then the Powell algorithm applies the searched parameters as its initial values and approaches for the approximate optimal value in local area. Experimental results show that improved algorithm decreases the registration error, improves the accuracy, and achieves the sub-pixel level in image registration.

Key words: Powell algorithm, normalized mutual information, genetic algorithm