The purpose of image-map registration is to revise the digital map included in Geographic Information System (GIS) with aerial image. The traditional method for this task requires the manual selection of features such as tie points in both the image and the map. Determining how to automatically extract common features between image and map is a difficult problem. In this study, we propose a novel framework of extracting the common feature named Diffusion Geodesic Path (DGP) between image and map, and use it for image-map registration. In order to overcome the differences in presentations between image and map, we first construct two new edge images from the image and the map using the diffusion process in physics to diffuse the differences between them. Then, we extract several paths between feature points (corner points) on these new edge images using the geodesic. Finally, a part of the extracted paths are automatically selected and are used for registration. The experimental results implicate that our DGP is a common and robust feature between image and map, and it can provide sufficient information for automatic image-map registration. The whole framework works well on automatic image-map registration.