A novel approach for face verification on mobile devices

Abstract

Most people have cellular phones nowadays. In this technological age, much attention had been given to information security. Many approaches and methods had been developed. Some systems such as security codes, fingerprint or signature verification are used to secure these devices. Processors used in such devices have less computational power than PC processors. On the other hand, high resolution cameras, i.e. cameras with 8 mega pixel, have been attached to these handheld devices that enable face verification algorithms to have better accuracy. Despite of such higher resolution cameras, mobile devices processors are not able to process large face verification in real time. Therefore, face verification algorithms need to be modified to enable hand phones to verify the owner in acceptable time. This project focuses on face verification technique to provide a reliable identification method. Principle Component Analysis and Gabor-Adaboost methods were simulated. The strength and weakness of the methods were investigated further. The comparison was based on the accuracy of verifications and the time taken for each simulation. Furthermore, this project intends to seek a robust face verification system. The algorithms were implemented to ORL face database and Essex face databases (Face 94 & Face95). The results showed that Gabor-Adaboost method is relatively more robust than Principle Component Analysis with regards to pose variation and illumination changes, by having more than 95% of verification accuracy.