Data Augmentation and Feature Fusion for Melanoma Detection with Content Based Image Classification

Abstract

Computer aided diagnosis has leveraged a new horizon for accurate diagnosis of numerous fatal diseases. Melanoma is considered as one of the most lethal form of skin cancer which is increasingly affecting the population in recent times. The disease can be completely healed if diagnosed and addressed at an early stage. However, in most of the cases patients receive delayed care which results in fatal consequences. The authors have attempted to design an automated melanoma detection system in this work by means of content based image classification. Extraction of content based descriptors can nullify the requirement for manual annotation of the dermoscopic images which consumes considerable time and effort. The work has also undertaken a fusion based approach for feature combination for evaluating classification performances of hybrid architecture. The results have outclassed the state-of-the-art outcomes and have established significant performance improvement.