Master of Web Science

Abstract

Nowadays, the number of websites is growing rapidly coupled with the hacking of the accounts which based on the traditional login method that depends on username and password, so it is important to improve the security of these sites by developing a method of verifying the identity of the users, we propose a face recognition system to achieve this target. In this paper, we review a method to detect faces and eyes depending on skin color then depending on Viola-Jones algorithm. After that, we propose our system which detects faces, eyes, and glasses by using the combination of skin color and Viola-Jones, then we normalize the
detected face image. After that we select facial features and configure a face template. This template is stored in the case of registering a new user or compared with pre-stored templates in the case of login. Our experiment reveals that the Detection of 110 images from the FERET database provides 100%, 90% and 69% accuracy in terms of face Detection, eyes Detection and glasses Detection, respectively. The recognition of 80 images from the ORL database (two images for each user) using 320 images related to 40 users (eight for each user) for algorithm training provides 93.75%, 92.5% and 88.75% accuracy in terms of using Eigenface, Fisherface and LPP, respectively.