Body Localization of an Arbitrary Skin Image

Research interest: All body regions naturally present different skin and hair patterns. Depending on the size of
a skin image, it can become very difficult, also for dermatologists, to recognize the body region, even when these
patterns remain visible. Using machine learning, we try to find discriminating visual features, which would allow to
determine the localization of skin images on the body.

Applications: Skin diseases often appear in specific body regions making the location of a skin lesion an important
feature in the differential diagnosis process. If the location is known, we can filter out improbable skin diseases and
refine our system predictions. In clinical settings, the localization prediction can then be trivially validated. Another
possible use for our model is in teledermatology applications where the uploaded patient pictures could be checked
to make sure they comprise specific body regions. We also envisage the creation of a tool, which would enable users
to filter public dermatology databases for specific body regions.