Software Based Biometric Liveness Detection using Convolutional Neural Network

Identifying a person based on some set of unique features is an important task. The human identification is possible with several biometric systems in which fingerprint and sclera recognition are the promising ones because of their individuality, uniqueness and reliability. A fingerprint image consists of a pattern of the valleys& ridges on human fingertips. The sclera is the white portion in the eye. The vein pattern seen in sclera and minutiae points in fingerprints is unique to each person. However the biometrics can be easily spoofed using several means. Pre-trained networks such as convolutional neural networks (CNN) can be explored for spoof biometric detection purpose. CNNs can achieve state-of-the-art performance even by training with natural images (such as animals, car, people etc. ). A software based approach is adopted in the work in which fake traits can be identified once the images are loaded and processed using software. Dataset Augmentation, process of increasing dataset, can be used to increase the classifiers (Support Vector Machine) performance. Single classifier is trained using all available dataset for improved accuracy and robustness.