Hardware acceleration of SVM-based classifier for melanoma images

Abstract

Melanoma is the most aggressive form of skin cancer which is responsible for the majority of skin cancer related deaths. Recently, image-based Computer Aided Diagnosis (CAD) systems are being increasingly used to help skin cancer specialists in detecting melanoma lesions early, and consequently reduce mortality rates. In this paper, we implement the most compute-intensive classification stage in the CAD onto FPGA, aiming to achieve acceleration of the system for deploying as an embedded device. A hardware/software co-design approach was proposed for implementing the Support Vector Machine (SVM) classifier for classifying melanoma images online in real-time. The hybrid Zynq platform was used for implementing the proposed architecture of the SVM classifier designed using the High Level Synthesis design methodology. The implemented SVM classification system on Zynq demonstrated high performance with low resources utilization and power consumption, meeting several embedded systems constraints.