Matching Face Images for Biometric Authentication

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 face images, using convolutional neural networks, directly from the raw pixel values of the input image. As the neural network is used for matching instead of classifications, modifying the users that the proposed method can recognize is a matter of adding or removing model images of the user's faces, without the need to retrain the model. The similarity between the face image and every model image is measured in order to select the user with the highest similarity to the input image as the recognized user, where that similarity measure is compared to a threshold value in order to authenticate that user. The evaluation results of the proposed method, using ORL shows that the proposed method has 99.48% accuracy with 0.22% False Acceptance Rate (FAR) and 3.2% 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.