基于局部结构张量-互信息的多模态图像配准

摘要

Mutual information ( MI) measure considers the global characteristics of image gray statistics only, and ignores spatial structure information and local characteristics of image gray statistics. In order to overcome these drawbacks, a registration method on the basis of the new measure of local structure tensor-mutual information ( LST-MI) is proposed. The proposed LST-MI measure considers the structure information of image neighborhood fully, gives the pixel position with greater importance larger weighting factor. Thus, the distinguishing of global extremum strengthens, the risk of trapping at local extremum reduces, the success rate improves, and the robustness of regis-tration enhances. Moreover, some registration experiments are conducted on simulated brain images and clinical im-ages. The results show that, in comparison with the registration method on the basis of mutual information and local mutual information, the proposed method improves the success rate of registration by more than 50%, and enhances the registration robustness significantly.%互信息测度仅考虑了图像全局一致的灰度统计特性而忽略了空间结构信息和图像灰度统计的局部性特点.为克服以上缺点,提出了一种基于局部结构张量-互信息新测度的配准方法.所提出的新测度充分考虑了图像邻域的结构信息,将结构越强的位置赋予较大的度量值,使得全局极值的区别性加强,减少了配准过程中陷入局部极值的风险,提高了配准成功率,增强了配准的鲁棒性.采用模拟脑图像和临床图像进行了配准试验,结果表明,与基于互信息和局部互信息的配准方法相比,该方法配准成功率平均提高了50%以上,配准鲁棒性显著增强.