Multimodal Image Matching based on Multi-scale Structure Feature

Abstract:
To solve the situation that the imaging features of SAR and visible images are widely different in scene matching, a multimodal image matching method using multi-scale structural features is proposed. This method introduces the concept of graph structure, transforms the multimodal image matching problem into a weighted bipartite graph matching problem, extracts the important edge screening nodes from the node attribute similarity and location similarity, constructs a similarity matrix between the template and the real-time image, and obtains the similarity of the two graph structures by solving the matrix. After achieving the extraction of interest regions for graph structure matching, fine matching is performed with LBP texture histogram to finally achieve the precise location of multimodal images. The experiment proves the effectiveness of our method for visible light and SAR image matching is better than other two classical methods.