Vgg16 based Transfer Learning and SVM based Classification for Identifying Gender from Human Iris
Mayank Rajput, Ganesh S. Sable · International journal of imaging and robotics · 2020
This research is motivated by some important research questions related to existing biometric systems. Existing biometrics either accept the person if enrolled within a database of the system or reject if not registered. They do not reveal other information about the person, such as age, gender, and ethnicity, etc. The high-security organizations such as banks, airports, border control would be at the risk if impostor tries to get an entry in it, but get rejected without knowing any other description of that individual. In the field of a criminal investigation, knowing the gender and age of a criminal will be helpful as it will reduce search time. In this context, the proposed system identifies the gender from iris biometrics. Experimentations are performed on two different databases GMBAMU-IRIS database and ND-GFI database. The proposed method implements a Vgg16 pre-trained neural network for transfer learning by extracting features from its pool5 layer. Further multi-class SVM is employed for classifying gender from features. The proposed work obtained promising accuracy for both databases. It also validates the hypothesis stating that the human iris consists of gender related features.