This document discusses a hierarchical classification system for skin cancer images. It uses binary classifiers and voting to achieve higher accuracy (76%) than a pretrained CNN model (22%) with fewer images. The system initially classifies lesions with CNNs. Features are extracted from the CNN output and used to build a hierarchical classifier with pairwise classifiers and voting. This approach derives a taxonomy of skin cancer classes from image data alone, replicating expert domain knowledge. It can be deployed on mobile applications to improve healthcare access.