一种结合梯度方向互信息和多分辨混合优化的多模图像配准方法

A Multimodal Image Registration Method Combining Gradient Orientation Mutual Information with Multi-resolution Hybrid Optimization Algorithm

针对基于传统互信息图像配准容易产生局部极大值,同时结合梯度信息的互信息改进方法不能很好地应用于梯度幅值差异较大的多模图像配准,提出了一种新的结合梯度方向的互信息测度函数.在参量优化过程中,将具有全局优化的遗传算法和Powell局部优化算法动态结合,前者的配准结果为后者的算法优化提供有效的初始点以抑制局部极值,同时借鉴小波变换中多分辨率的思想,在低分辨率图像中粗略配准后,上升到高分辨率图像上进一步细化配准结果,增加算法鲁棒性并减少优化时间.多幅红外与可见光图像配准实验结果证明,提出的算法具有配准精度高和鲁棒性强等特点.

In order to solve the problem that the image registration method based on the classical mutual information may suffer from local maxima and these improved methods combing with gradient can not align the multimodal images because of the great difference on gradient amplitude,a novel multimodal image registration method is proposed.During the optimization of transform parameters,a hybrid optimization algorithm based on genetic algorithm（GA） and Powell is carried out to efficiently restrain local maxima of this new similarity measure function.The former provides the latter with effective initialization parameter,which will increase the algorithm＇s robustness.Multi-resolution data structure based on wavelet transform is used to expedite the registration process.Experimental results demonstrate that this new algorithm can efficiently speed up the registration process with a good registration result.