I. INTRODUCTION
Skin cancer represents a serious public health problem because of its increasing incidence and subsequent mortality. Among skin cancers, malignant melanoma is by far the most deadly form. As the early detection of melanoma significantly increases the survival rate of the patient, several non-invasive imaging techniques, such as dermoscopy, have been developed to aid the screening process [1]. Dermoscopy involves the use of an optical instrument paired with a powerful lighting system, allowing the examination of skin lesions in a higher magnification. Therefore, dermoscopic images provide a more detailed view of the morphological structures and patterns as compared to the normally magnified images of the skin lesions [1], [2]. However, the visual interpretation and examination of dermoscopic images can be a time consuming task and, as shown by Kittler et al. [3], the diagnosis accuracy of dermoscopy significantly depends on the experience of the dermatologists. Several medical diagnosis procedures have been introduced in order to guide dermatologists, such as the pattern analysis, the ABCD rule, the 7-point checklist, and the Menzies method. Overall, in these methods several dermoscopic criteria (i.e. asymmetry, border, colors, differential structures) have to be assessed to produce the final clinical diagnosis. However, the diagnosis of skin lesions is still a challenging task even using these medical procedures mainly attending to the subjectivity of clinical interpretation and lacking of reproducibility [1], [2].