Deep Color Spaces for Fingerphoto Presentation Attack Detection in Mobile Devices

Abstract

AbstractFingerphotos are fingerprint images acquired using a basic smartphone camera. Although significant progress has been made in matching fingerphotos, the security of these authentication mechanisms is challenged by presentation attacks (PAs). A presentation attack can subvert a biometric system by using simple tools such as a printout or a photograph displayed on a device. The goal of this research is to improve the performance of fingerphoto presentation attack detection (PAD) algorithms by exploring the effectiveness of deep representations derived from various color spaces. For each color space, different convolutional neural networks (CNNs) are trained and the most accurate is selected. The individual scores output by the selected CNNs are combined to yield the final decision. Experiments were carried out on the IIITD Smartphone Fingerphoto Database, and results demonstrate that integrating various color spaces, including the commonly used RGB, outperforms the existing fingerphoto PAD algorithms.KeywordsColorspacePresentation attack detectionFingerPhoto