Abstract: The traditional image matching algorithm based on spectral graph usually matches the points with the position relationship of feature points，and the gray information around feature points is not fully utilized． In order to solve this problem，this paper proposes an image matching algorithm based on spectral features． This algorithm uses the spectrum of line graph to reflect the changes of the gray level around feature points，stratifies the neighbors of each feature point，and then constructs a line graph for the points of each layer． Thus，the spectral features of feature points are obtained from the spectrum of line graph． Theoretical analysis demonstrates that the spectral features are of rotation invariance，linear brightness variation invariance and strong robustness to noise． Finally，the Hungarian algorithm is used to solve the matching problem and output the matching results． Experimental results show that the proposed algorithm has a high matching accuracy，and it can also achieve better matching results under a
larger deformation between the two images to be matched．

Key words: image matching, local feature, feature description, line graph