Towards Fully Automated Face Verification
Ali Murtaza, Qamar Sarfraz, Syed Ahmed Afzal, Muhammad Ibrahim Syed, Khurram Khan, Zahid Mahmood · 2021
Automatic face verification under uncontrolled environment is a challenging task due to various crucial applications, such as surveillance, access control, and security. Despite the immense research in the past few decades, it is still an open issue due to factors, such as non-uniform illuminations, pose variation, low-resolution, expression, and occlusion. Owing to the aforementioned issues, this paper presents an adhoc study on three different poses and occlusion towards automated face verification. This study comprises of two face detectors, which are Dual Shot Face Detector (DSFD) and the Viola Jones algorithm. The verification methods investigated in this study are the AdaBoost-LDA, the LBP, and the PCA based algorithm. Simulations on face detection reveal that the DSFD method yields higher accuracy at the cost of higher computations. Whereas, face verification experiments reveal that both the PCA and LBP perform well under occlusion and pose variation than the AdaBoost-LDA. The AdaBoost-LDA is computationally efficient than the compared methods. In addition, all the three verification algorithms are near real-time.