Automatic Robust Registration of Cutaneous Hemangiomas for Follow-up Examinations

Sebastian Zambanini, Georg Langs, Robert Sablatnig, Harald Maier 1Pattern Recognition and Image Processing Group, Vienna University of Technology, Austria 2Institute for Computer Graphics and Vision, Graz University of Technology, Austria 3Division of Special and Environmental Dermatology, Medical University of Vienna, Austria {zamba, langs, sab}@prip.tuwien.ac.at, [email protected] Abstract: This paper presents an automatic method for registering follow-up hemangioma images taken during clinical trials in specific time intervals. The registration is a prerequisite for the analysis of subtle surface changes on the hemangioma, that requires knowledge about the previous state of local appearance to determine the development of the disease. The registration method has to handle the partially changing shape of the hemangioma during its healing process, while typically no surrounding landmarks are available to aid the registration process. The method finds interest points in two images and matches corresponding interest points using SIFT features to determine a homography between the two images by means of RANSAC. Experimental evaluation of the robust registration are reported for image pairs acquired at the same time point and during follow-up.