An Improved Methodology for Fraud Detection in Face Recognition Systems
In this study, a method has been developed in response to the tricking movement which is a problem encountered in biometric systems. A method for more reliable applications has been developed by performing liveness analysis in face recognition systems. A security measure has been taken by following the movements performed during normal operations without understanding that the users are being followed. As an image processing library, the Dlib library, an open source library with machine learning, deep learning and computer vision algorithms, has been used. This library contains a detector which can find faces in an image. With this detector, we can predict the face area of the person standing in front of the camera. Then, from the face area obtained, we obtain two eye regions. The iris area in the eye is determined by moving iris kernel matrix over the eye images. After determining iris, the middle point is found and the eye is monitored. Users can watch the manuscripts by reading dynamically changing texts on the screen in a certain time without notice to the iris movements. As a result of all these results, liveness decisions are made by making comparisons according to the determined threshold values. This algorithm has been tested in volunteers. The volunteers selected to take into consideration the various situations are composed of both young and old individuals. As a result, the algorithm was tested on the volunteers and it can give liveness decision correctly and the success rate was 97%.
Keywords - Biometrics, Spoofing, Liveness Detection, Dlib, Facial Landmark