Bilkent University
Department of Computer Engineering


Face presentation attack detection


Sepehr Nourmohammadi
Master Student
(Supervisor: Asst. Prof. Shervin Rahimzadeh Arashloo )
Computer Engineering Department
Bilkent University

Abstract: In the contemporary era, biometrics technology has become a highlighted area and center of attention due to conceivable progress of technology. The reason behind this is that biometrics cases might be applied in security in various aspects such as criminal investigation, provide financial and society immunity, etc. Because of the sensitivity of the functionality of face biometric systems to presentation attacks, vandalism approaches take place in order to penetrate to the individual or victim’s system and take advantage of them by using artifact biometric features of the system owner. due to the potential of the security risks, face attack detection obtained an enormous amount of attention during decades. In this research one-class classification (OCC) method is considered rather than binary classification which has been done by many researchers. The aim of OCC is providing a way to distinguish the positive labels or bonafide cases by disregarding the different types of attacks. This method brought an inevitable amount of interest in biometric communities, and other interdisciplinary fields of computer science. In each frame different areas of faces will be cropped and normalized, then transfer learning techniques are used to extract features. The driven features would be applied to multiple one-class classifiers. In order to compensate for misclassified cases, the ensemble learning method is going to improve the score results. In this research. instead of using traditional ensemble learning approaches to update the weight factor of combined scores such as weight bagging which leads to an overfitting, sum or product rule of combination will be applied, to update the weight factor of the combined scores Lagrange multiplier optimization theory is going to be considered that paves the way of boosting the performance while dealing with non-linear programming problems. Lagrange multiplier optimization theory is an innovative part that has not been considered in any research circulating around this field.


DATE: 04 April 2022, Monday @ 16:20 Zoom