Liveness detection in facial biometrics using complete dynamic local ternary pattern technique

Facial biometric systems have recently received increased deployment in various applications such as surveillance, access control and forensic investigations. However, facial biometrics facing various tangible threats, one of them is spoofing attacks. A spoofing attack occurs when a person tries to masquerade as someone else by non-real faces such as photograph, video clips or dummy faces and thereby gaining advantages from applications. In order to identify the spoofing attacks on such biometric systems, face liveness detection countermeasure have been developed.There are numerous ways to detect the liveness of face such as through motion analysis, texture analysis, identify life sign clues and thermal sensors. Recently, texture analysis has received more attention because of its nonintrusiveness;high efficiency and accuracy to discriminate face skin texture from spoof attacks. For this purpose, a numbers of texture descriptors have been proposed in the literature for face liveness detection. However, they exhibit some limitations in terms of noise with center pixel, manual setting of threshold (τ) value and ignorance of global intensity in the image.Thus, a robust face liveness detection method based on Complete Dynamic Local Ternary Pattern (CDLTP) has been proposed in this thesis. The CDLTP was designed to overcome the limitations of reported texture descriptors.Weber’s law was used to automatically set the threshold value in ternary pattern by considering the sign, magnitude and global intensity of the image.Its effectiveness has been tested and benchmarked against other existing texture descriptors (i.e. Local Binary Pattern (LBP), Local Ternary Pattern (LTP), Dynamic Local Ternary Pattern (DLTP), Complete Local Binary Pattern (CLBP) and Complete Local Ternary Pattern (CLTP)) on self collected database and other publically available databases (i.e. NUAA, CASIA and REPLAY-ATTACK).The evaluations have been carried out via statistical hypothesis testing and through liveness detection itself. The results have consistently demonstrated that CDLTP outperforms other techniques across various types of spoof mediums. The comparison analysis of CDLTP with other texture descriptors were also carried out on self collected and public domain face spoof databases. In all these experiments, the results obtained with CDLTP exceeds from the state-of-art.Various score level fusion strategies have been adopted to evaluate the performances of the overall system which comprises both face liveness detection and face recognition systems. The achieved decisions from scores level fusion strategies proved that CDLTP based face liveness detection reduces 89% of vulnerability of face verification system against spoof attacks.The measured result of serial method in which the face liveness detection performed before face recognition system was found to be the most effective methods among other score level fusion strategies that were analyzed