Techniques for a structural analysis of dermatoscopic imagery

Abstract

Techniques were developed for automated detection and characterization of dermatoscopic structures, including the pigment network and brown globules. These techniques incorporate algorithms for grayscale shape extraction based on differential geometry developed by Steger, a snake algorithm, and a modification of the region competition strategy of Zhu and Yuille. A novel approach was developed for global segmentation of pigmented lesions, based on stabilized inverse diffusion equations. Procedures for detection of air bubbles and hairs in dermatoscopic images are also reported.