Polarization Imaging for Face Spoofing Detection: Identification of Black Ethnical Group

Abstract

Recently, several studies have shown the ability of polarized light as one of the face spoofing countermeasures. In this paper, polarized light is used to identify genuine human user from black ethnical skin color. Printed photos are used as spoofing attacks. Then, the Stokes parameters are applied to generate I SDOLP image for each genuine face and printed photo. Visually, the I SDOLP of genuine black users seem brighter than the other skin colors. The mean intensity has erroneously classified all the I SDOLP images of black skins as photo faces. Coarsely comparing I SDOLP histograms of black skin and printed photos shows that data distributions between the black skin and printed photo are relatively similar. The bimodality coefficient (BC) algorithm is then applied to study the distributions modality. Surprisingly, the BC has been able to identify these genuine black skin group well, but fails to other ethnical groups. A newly proposed fusion formula which is named as the Mean_BC (MBC) has achieved higher detection accuracy rate and can be a robust face spoofing detection measure for polarized database consists of various ethnical groups.