Development of deep learning-based 3D face fraud prevention system for face recognition systems

Biometrics refers to the processes used to verify, recognize and identify people based on their physical or behavioral characteristics. Physiological features that can be obtained for biometric identifiers include fingerprints, palm veins, palm prints, face, DNA, iris, retina and hand geometry. Face recognition systems, which are the most popular biometric identifiers, basically serve as an automatic authentication system by detecting the facial regions in the images using machine learning algorithms and matching them with the people registered in the system before. Face recognition technologies provide many advantages over other technologies such as iris and fingerprint recognition and biometric identifiers as they are convenient, contactless and easy to apply. Unfortunately, the advantages of biometrics turn into disadvantages when a biometric data is stolen or copied. This risk is much higher than other biometric data in face recognition systems where one or more photos/videos of a registered user can be obtained easily and cheaply over the internet or only the face is captured using a camera even without the user's consent or physical contact. Face recognition systems supported by liveness detection algorithms are seen as a very important and critical need in the security sector. The aim of this thesis is to develop artificial intelligence technologies and deep learning algorithms and face fraud detection algorithms that will maximize the reliability and accuracy of face recognition systems. When the results of the applications are examined, a stable, high-performance and high-accuracy model, which is not affected by attacks to deceive face recognition systems, has been obtained with a success performance of 99.5%.