关键词: 互信息, 图像配准, 高阶熵

Abstract: Mutual information has been widely used in image registration as an effective similarity measure. However, standard mutual information only considers the pixels’ intensity value but ignores their spatial relationship. Therefore, the lack of spatial information leads to the poor robustness of the method. In this paper, the mean value of each pixel’s neighborhood has been incorporated as the second term of the high-order entropy. Thus, the spatial information can be successfully added into the mutual information. The experimental results show that our method has good performance against noise compared with the standard mutual information. The smoothness of the curve for registration function can avoid the searching trapping into the local minimum which often leads the registration to failure.

Key words: mutual information, image registration, high-order entropy