Melanoma rates in whites are increasing rapidly. Early detection, primarily through altered appearance of pigmented skin lesions, is the strongest determinant of survival. For high-risk patients photographic records are invaluable, but images must first be "registered" to remove distortions introduced during imaging. This paper describes an algorithm for registering digital images of multiple skin lesions automatically. The algorithm comprises four stages: (1) generate lesion maps; (2) identify initial matching lesions;(3) select four matching corner-lesions; and (4) transform one image. Performance tests using images of both simulated and actual skin lesions show that matching and registration are successful under realistic imaging conditions. (C) 1998 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved. [References: 38]
展开▼
