Smartphone-based fingerprint authentication using siamese neural networks with ridge flow attention mechanism

Abstract

Authenticating finger photo images captured using a smartphone camera provides a good alternative solution in place of the traditional method-based sensors. This paper introduces a novel approach to enhancing fingerprint authentication by leveraging images captured via a mobile camera. The method employs a siamese neural network (SNN) combined with a ridge flow attention mechanism and convolutional neural networks (CNN). Our approach begins with collecting a dataset consisting of finger images from two individuals then we apply multiple preprocessing techniques to extract fingerprint images, followed by generating augmented data to improve model robustness, scaling, and normalizing them to form images suitable for model training. Next, we generate positive and negative pairs for training a SNN. We used the SNN with CNN for feature extraction, combined with an attention mechanism that focuses on the ridge flow pattern of fingerprints to improve feature relevance which significantly contributed to the performance enhancement. As for the testing performance, our model has an accuracy of 90%, precision of 89%, recall of 83%, F1 score of 86%, area under the curve (AUC) 95 %, and 13% of equal error rate (EER) when using smartphone-captured images for fingerprint recognition.