Abnormal Cervical Cell Detection using HoG Descriptor and SVM Classifier
Sophea Prum, Dini Oktarina Dwi Handayani, Patrice Boursier · 2018
Cervical cancer is preventable and treatable if it can be detected at early stage by using simple Papanicolaou (PAP) test. Unfortunately, the incident rate in developing countries is still high due to lack of national PAP test policy and limited number of trained cytotechnologists and cytopathologists. Therefore, using image processing system to analysis PAP screening image can be an effective solution. We proposed a machine learning based system for abnormal cell detection as well as cell type classification. The proposed system relies on HoG feature extraction method and SVM classifier. The experiment results conducted on Harlev dataset have shown a recognition rate of 94.70% for normal and abnormal cell classification when considering that nucleus in each cell image is perfectly detected. When considering a real scenario application by using our proposed nucleus detection method, our system archives 88.83% of recognition rate.