Face presentation attack detection using multi-classifier fusion of off-the-shelf deep features

Abstract

Face recognition systems are vulnerable to the presentation (or spoof or direct) attacks that can be carried out by presenting the face artefact corresponding to the legitimate user. Thus, it is essential to develop a Presentation Attack Detection (PAD) algorithms that can automatically detect the presentation attacks the face recognition systems. In this paper, we present a novel method for face presentation attack detection based on the multi-classifier fusion of deep features that are computed using the off-the-shelf pre-trained deep Convolutional Neural Network (CNN) architecture based on AlexNet. Extracted features are compared using softmax and Spectral Regression Kernel Discriminant Analysis (SRKDA) classifiers to obtain the comparison scores that are combined using a weighted sum rule. Extensive experiments are carried out on the publicly available OULU-NPU database and performance of the proposed method is benchmarked with fifteen different state-of-the-art techniques. Obtained results have indicated the outstanding performance of the proposed method on OULU-NPU database.