Real Time Face Recognition Comparison Using Fisherfaces and Local Binary Pattern

Abstract

Face recognition has been gaining popularity by computer vision researchers over last two decades. Face recognition concerns to identify person from an image set. In general there are three face recognition classes, i.e., holistic based, feature-based, and hybrid methods. Fisherfaces extends Eigenfaces approach using Fisher's linear discriminant to improve classification rate by maximising the ratio of between-class to within-class scatters. In the other hand, local binary pattern employs shape and texture in local pixel neighbourhoods to build a global representation of a face image. Both methods are implemented and analysed using three distinct face image dataset and a real time video application. The accuracy, training and testing time for both algorithms are measured in some experiments using k-fold cross validation scheme. From experiment result, it is concluded that Fisherfaces has a good prediction time that makes it a good choice for real time face recognition applications. In contrast, local binary pattern can handle classifier addition, so it can be used in dynamic face recognition scenarios.