Skin disorder diagnosis with ambiguity reduction assisted by lesion color adaptation

A smart phone application based on a low complexity image processing technique and a novel fuzzy-like classification method are presented for skin disorder diagnosis. The proposed classification method takes into consideration the size and color features of skin lesions rather than their shape and texture. The classification rules are determined after processing statistically a small number of representative training photographs. Consequently, they can be defined by an end user that is not necessarily skilled in computer science. The application presented in this paper can serve as a complementary tool for a dermatologist to continuously monitor remotely his patients. The accuracy of the diagnosis that is based only on the image processing outcomes, ranges between 85.3% and 97.7% using 5 only representative photographs as a "training set" (corresponding from 9% to 24% of the test set per disease). The achieved accuracy can be improved (up to 17%), if the photographs are processed using a specific color adaptation technique. The small fraction of training photographs can be scaled up if the size of the test set is increased but it is expected that a limited number of training photographs will be sufficient in order to achieve an acceptable accuracy for a test set of any size. This accuracy can be further improved if other factors are taken into consideration (progression of the symptoms, information provided by the user, etc).