Abstract: A new method is posed that analyzes the similarity between the 2D images based on the landmark positions. Get the coordinates of the landmarks and standardized them to make up two groups of vectors, and analyzed the correlation between them, which represented the similarity of the two images. Five cephalometric radiographs images were analyzed in this method and factors affecting the similarity were discussed. It indicated that the number of landmarks and the method of selection could influence the results remarkably but random error had little influence. This method can evaluate the similarity between the images quantitatively and can also be applied to the 3D images.

Key words: canonical correlation analysis, medical image, cephalometric radiographs