Matching Fingerprint Images for Biometric Authentication using Convolutional Neural Networks

The use of biometric features, to authenticate users of different applications, is growing rapidly in recent years, according to the high sensitivity of the protected information and the good security that biometric authentication provides. In this study, a method is proposed to measure the similarity between two fingerprint images, using convolutional neural networks, instead of classifying them. Thus, modifying the users that the proposed method can recognize is a matter of adding or removing model images of the users' fingerprints. The similarity between the fingerprint image and every model image was measured in order to select the user with the highest similarity to the input image as the recognized user, where that similarity measure was compared to a threshold value in order to authenticate that user. The evaluation results of the proposed method, using FVC2002_DB1 and FVC2004_DB1 showed that the proposed method had 99.97% accuracy with 0.035% False Acceptance Rate (FAR) and 0% False Rejection Rate (FRR). Hence, the proposed method has been able to maintain high accuracy while eliminating the vulnerabilities of biometric authentication systems imposed by the use of separate stages for features extraction and similarity measurement.