Secure Face Recognition Model Employing Image Quality Assessment for Anti-Spoofing

Abstract

Face recognition technologies have long been incorporated into various advanced security and commerce systems. However, they are still vulnerable to sophisticated presentation attacks. Therefore, a secure face recognition system architecture is proposed that combines quality and anti-spoofing controls. This closes the possibility of identity fraud. The particular architecture under consideration takes into account estimations of several intrinsic quality factors such as uniformity of illumination, structural uniformity, edge sharpness, textural roughness, background noise, etc. applying such factors to differentiate between genuine face images and the fraudulent ones, including photographic prints, video recordings, and digital forgeries. The architecture incorporates those features with face images subjected to deep convolution training to enhance the performance of a discriminative classifier most suited to withstand varied attacks. The architecture has been tested against a number of prominent datasets and the findings demonstrate the improvement of the detection accuracy along with the improvement of the attack success rate, as well as the system performance stability across different lighting and background capture conditions. The findings indicate the use of IQA as the last line of defense is what is currently required by face recognition technology, and the addition of IQA as one more layer greatly enhances the face recognition technology.