Fingerprint Presentation Attack Detection Algorithm

Given the growing interest in security systems, researchers and developers are continually devising new methods of using biometrics, in particular fingerprints. However, hackers have discovered a flaw in the fingerprint method and exploit it through presentation attacks (PA). As a result, there is a need to protect the safety and security of authentication systems against PA through automatic presentation attack detection (PAD). The current research showed promising results after the application of PAD to deep learning (DL) algorithms. The proposed model is a softwarebased approach, namely a binary classification system to distinguish between live and fake fingerprint images as these were the most appropriate in terms of cost. In order to achieve the goal of this research, we applied a three-layer convoluional neural network (CNN) using the ATVS-FFp DB dataset for training and testing. Two different splits of the data have been tested in this worksThe first time, the proportions were training =60%, valid =20%, and testing =20%, and in the second division, we reduced the images for verification, increased them in the test, and kept the training rate as it was, namely the valid =10%, and testing =30%. The model showed that it is affected by the number of images in the test and shows higher results the more the test phase is given. The trained model achieved 99% training accuracy and 97% testing accuracy.