Hierarchical Content-Based Image Retrieval of Skin Lesions

This paper proposes a novel hierarchical content-based image retrieval system and its application to skin lesion images. Five common classes of skin lesions, including two non-melanoma cancer types, are used. Colour and texture features are extracted from lesions. Feature selection is embedded in a hierarchical framework that chooses the most relevant feature subsets by comparing different similarity for each level of the hierarchy. Experiments on our database of 533 images show that the proposed hierarchical scheme improves retrieval precision by about 4%, reaching a maximum average precision of 78%.