Study on Local Maxima Problem of Mutual Information in Different Resolution Images Registration
图像插值方法对互信息局部极值的影响分析

Mutual information has been used as a similarity metric in medical image registration. The probabilities of mutual information may be estimated by normalization of the joint intensity histogram, which is obtained by binning the intensity pair of the overlapping parts of the reference image and the floating image. However, image interpolation would create new intensity pairs and may cause local maxima of mutual information. In this paper, local maxima of mutual information are analyzed using linear interpolation and near neighborhood interpolation for different resolution images. Analysis results show that mutual information contains less local maxima when linear interpolation is used, and contains local maxima when near neighborhood interpolation is used. Experiments show the validity of the results. All these results are benefit to multimodal medical image registration.