Convolutional Neural Networks for Fingerprint Liveness Detection System

Fingerprint liveness detection is a challenge that possess tremendous security risks. The detection is a technique for determining the liveness of a given biometric system and to ensure it is not fake. Fingerprint attack deals with images coming from artificial replication of symptomatic fingerprints, which are made of wood glue, gelatin or latex. To obtain a better liveness detection system, a convolution neural network is used to develop a system in our work. Our system is evaluated with LivDet 2019 Database which consist of 2000 live and fake biometric fingerprint images. Within the small training set containing 1300 samples, the trained model achieved 100% training accuracy and testing accuracy of 95%. The features taken from the CNN were given to machine learning classifiers for a comparison study where K-Nearest Neighbors, Logistic Regression and Naïve- Bayes classifiers gave results as good as the CNN architecture.