Aiming at the problem of multi-sensor image matching when the light intensity is too high or too low, this paper proposes a method of multi-sensor image matching based on joint graph spectrum feature analysis. Firstly, the K-nearest neighbor rule is used to calculate the structural relationship between the corner points in the visible light and infrared images to construct a joint graph. Then, based on Laplacian decomposition, we calculate the eigenvalues of the adjacency matrix in the joint graph to obtain the fea- ture vector of the joint graph, and construct the feature function pair through three-dimensional reconstruction. Thirdly, a maximally stable extremal regions detector based on SUSAN-MSER-SURF is proposed to detect the extreme value position of the feature func- tion pair. Finally, by normalizing the maximum stable extremal regions and matching them, the exact matching results of multi-sen- sor images can be obtained. The experimental results show that the proposed matching method based on the spectral feature analysis of the joint graph can solve the problem of multi-sensor image matching when the light is too high or too low.