Fingerprint Liveness Detection with Voting Ensemble Classifier

Detecting a user's fingerprint is a common verification process in many daily products such as smartphones and laptops. The convenience makes it popular, but this method is vulnerable to a presentation attack. Any fingerprint can be copied onto materials such as wood glue and gelatin, using only a few simple steps. Therefore, detecting whether the fingerprint comes from a live person is essential. In this paper, we proposed a method that employs a voting ensemble classification model to aggregate predictions from multiple individually trained machine learning models to determine whether an input fingerprint image is a live or a fake one. The input image is first pre-processed with a wavelet denoising algorithm, then Local Binary Pattern (LBP) and Local Phase Quantization (LPQ) are used for feature extraction. Next, we train a Voting Ensemble Classifying Model, utilizing predictions from several trained models, to find the majority vote for fingerprint liveness detection. According to the performance on the public LivDet 2015 dataset, the proposed method achieved better accuracy classification error (ACE) compared to the state-of-the-art models on three out of four sensor types: Greenbit (ACE=0.95%), Digital Persona (ACE=3.71%), and Hi Scan (ACE=1.39%).