| Fisherface-based, Eigenface-based and Direct Correlation-based facial recognition methods are often applied approaches for facial recognition. As it is known, these methods are highly affected by noise, illumination and other artifacts. Even though these methods have been around for more than twenty years, what additional factors affect their accuracy have not been accordingly studied as of yet. The main contribution of this thesis is the analysis and the investigation of how and what affects the recognition accuracy of Eigenface-based, Fisherface-based and Direct Correlation-based methods. In particular we present a facial recognition system that applies these three facial recognition algorithms to various databases containing images that have been preprocessed and/or grouped in different ways based on some proposed criteria. We show the effects on recognition accuracy of the usage of registered vs. raw images, the size and selection of the training templates, and the selection of the amount and type of information contained in the face images. Finally experimental results are presented to demonstrate the potential value and importance of each of these proposed factors on facial recognition. |
