An Anti-Occlusion Matching Algorithm of Multimodal Images Based on Graph Structure

Occlusion is a common difficulty in application of multimodal image matching, which is essential for navigation and computer vision. However, recent multimodal image matching algorithms based on traditional methods lack research on anti-occlusion issues. To improve the matching probability of multimodal images with occluded target, in this paper, we proposed an anti-occlusion matching algorithm based on the graph structure by utilizing the consistency of the graph structure in the un-occluded areas. Firstly, we extracted corner point features from images and constructed them to graphs. Secondly, we proposed an improved graph matching algorithm to find corresponding points by measuring the similarity between local graph structures. Finally, we took advantage of a global mismatch elimination algorithm to obtain more robust results. The comparative experimental results on visible-infrared datasets demonstrated that the algorithm we proposed had improved matching probability, matching accuracy and had certain anti-occlusion ability.