Face spoof detection using light reflection in moderate to low lighting

Face recognition is widely viewed as an alternative means of authentication to replace traditional password methods in different applications for access control. Despite significant improvements, this form of authentication remains plagued by a number of vulnerabilities ranging from the use of printed photographs, 3D masks, and video replay attacks, prompting the need for a more robust approach in defending against such spoof attacks. Using the observation that different materials reflect light differently, we propose a system that uses light reflection patterns and night vision infrared to detect spoof attacks. We developed a system using Laplacian blur detection, Gabor filters, color moments and Local Binary Patterns, which calculates the reflection of light on different material and classifies whether the given face is real or fake. We noticed significant improvement in our results, with the system working well in a lighting controlled environment that is comparable to some existing systems. In particular, a single light source when capturing a sample for preprocessing yielded optimal results. Furthermore, we also created our own dataset comprising 40 individuals using several cameras that can serve as another source in addition to the existing CASIA-FASD and MSU MFSD public datasets.