RGB-T Cross-Modal Image Registration Fusing Global Structure and Local Mutual Information

Cross-modal image registration is a prerequisite for visual information fusion in target recognition and robot navigation. However, cross-modal image registration is challenging due to the significant differences in texture and grayscale of different modes of images. To solve this problem, this paper proposes an RGB and thermal (RGB-T) image registration method that combines global structure with local mutual information, called GSLMI. First, the mutual information (MI) in local optimal regions is utilized to calculate the initial value of transformation matrix. Second, global consistent edge structures are detected to obtain similar contents of RGB and thermal images. Third, feature points are extracted by the curvature scale space (CSS) corner detector based on global edges, and the main orientation of points are calculated by the contour angle orientation method. Finally, feature matching is performed with MI providing initial value guidance to achieve the image registration. The accuracy of the proposed method GSLMI is demonstrated by registration experiments on a self-constructed RGB-T images set. The results show that GSLMI can achieve superior performance over the state-of-the-art methods.