Online Face Recognition with Application to Proactive Augmented Reality

Abstract

Recently, more and more researchers have concentrated on the research of video-based face recognition. The topic of this thesis is online face recognition with application to proactive augmented reality. We intend to solve online single-image and multiple-image face recognition problems when the influence of illumination variations is introduced.

First, three machine learning approaches are utilized in single-image face recognition: PCA-based, 2DPCA-based, and SVM-based approaches. Illumination variations are big obstacles for face recognition. The next step in our approach therefore involves illumination normalization. Image preprocessing (AHE+RGIC) and invariant feature extraction (Eigenphases and LBP) methods are employed to compensate for illumination variations. Finally, in order to improve the recognition performance, we propose several novel algorithms to multiple-image face recognition which consider the multiple images as query data for subsequent classification. These algorithms are called MIK-NN, MMIK-NN and Kmeans+Muliple K-NN.

In conclusion, the simulation experiment results show that the LBP+x2-based method efficiently compensates for the illumination effect and MMIK-NN considerably improves the performance of online face recognition.