Face Recognition using Enhanced Versions of Principal Component Analysis for Feature Extraction

Abstract

PCA is more than 100 years old now. However, it still receives new enhancements and plays new roles in dimensionality reduction for many applications. This thesis assesses the current status of PCA and its enhancements for feature extraction for face recognition. The thesis has implemented PCA and several enhanced versions of it on ORL face database. The performance of these PCA algorithms is reported and analyzed. The modular weighted 2D2PCA shows the best performance in these experiments and has achieved 94% recognition rate. The thesis also tests the performance of integrating ImagePCA and modular weighted PCA as one feature extraction method. The integrated method attains the same recognition rate as of modular weighted PCA with fewer coefficients. Finally we have compared the performance of those techniques with only one image per person for training which dropped in recognition rate by 22%.