Minkowski广义距离与多模态图像配准

A Similarity Measure Based on Minkowski Generalized Divergence for Multimodal Image Registration

根据图像灰度的联合概率分布函数与图像相似程度之间的变化规律,分析了Shannon互信息与K u llback-L e ib ler距离之间的关系,利用变量间的不等式关系理论,提出基于M inkow sk i不等式的广义距离度量,并构造了基于这一距离的多模态图像配准新测度.新的配准测度不再要求概率分布必须满足连续性的要求,实验中使用M R和PET医学图像进行了实验分析.结果显示,基于M inkow sk i距离的新配准测度比传统的信息论测度具有更强的噪声鲁棒性,用乘方运算代替了对数运算,数学表达式更简单,并省去了除法运算,在算法上也更容易实现.

Based on the relationship between the joint probability distribution function of two images and the similarity between images, the connection between Shannon mutual information and Kullback-Leibler divergence is investigated. Thus a novel definition of divergence measure based on Minkowski inequality is proposed. On the proposed Minkowski generalized divergence the corresponding similarity measure for multimodal image registration is put forward. Unlike the information theoretic registration measures, Minkowski generalized divergence does not require that the condition of absolute continuity must be satisfied by the probability distributions involved. The new measure is applied to the rigid registration of clinical multimodal medical images. Experiment results show that the Minkowski similarity measure, when compared with information therotic measures, is more tolerable to noise and easier to implement in the Minkowski similartiy measure function clue to its simplicity, e.g. in the use of power operation instead of logarithmic computation avoiding division.