Face Spoofing Detection System using Local Invariant Feature Set

Face is a widely used biometric modality because of ease with which it can be captured by digital cameras. However, because of its wider accessibility and popularity, it becomes the most vulnerable biometric modality. These days spoofing attacks have become prevelant by which a fake user impersonates a genuine user. These attacks include photo attacks, video attacks and 3D mask attacks. For this a robust face spoofing detection system has been deployed consisting of handcrafted features. These features include rotation invariant Zernike moments, Polar Harmonic Transforms, discrete orthogonal moments consisting of Krawtchouk, Tchebichef and Dual-Hahn. The feature-set is invariant to rotation, scale and translation. These local features capture micro variations in an image that makes it possible for the system to differentiate between a genuine and a fake spoofed sample. For performance evaluation, publicly available databases: CASIA-FASD, REPLAY-ATTACK and OULU-NPU have been deployed. The proposed methodology shows an accuracy of 99.98% for CASIA-FASD, 99.98% for Replay-Attack and 99.95% for OULU-NPU databases. In case of OULU-NPU database, an ACER of as minimum as 1.95%, an APCER of 2.34% and BPCER of 1.56% is achieved. In addition to this, an EER of 0.063% and HTER of 0.165% is achieved for REPLAY-ATTACK database. For CASIA-FASD database, an EER of 1.698% is achieved. Thus, the proposed methodology shows a superior performance in comparison to techniques available in the literature survey.