Deep Learning with DCT for Liveliness Detection

Spoofing has grown to be a serious problem when digital payments are made for diverse transactions. In this work, we have addressed this problem using different deep learning models to distinguish between authentic and fraudulent fingerprints. Firstly, we have employed Discrete Cosine Transformation (DCT) to obtain the significant information from the given fingerprints. In the second phase, we have considered different deep learning models to obtain features which are subsequently used for classifying a given fingerprint as either authentic or fraudulent. In our work, we have explored LeNet, AlexNet and VGG models with DCT for fingerprint liveliness detection. The Support Vector Machine (SVM) is used as a classifier. Compared to the most recent findings, the proposed methodology yield better results.