Face recognition for biometrics-based personal identification

A thorough investigation in the area of face recognition is presented. We contemplate a variety of methods for the construction of feature spaces in face recognition. Linear and non-linear methods such as Eigenfaces, Fisherfaces, Isomap, and kernel-PCA are evaluated in terms of classification performance and robustness. The experimental environment comprehends well-known standards in the area of face recognition, namely YALE and FERET databases. Various lines of research are established in order to reveal the conditions in which the classifiers are likely to fail or thrive. Our investigations include:;The assessment of face classifiers in the presence of environmental disturbances. We include models of deterioration of visual information that mimic frequent scenarios in face recognition.;The evaluation of a modular approach to face recognition. We deliver an investigation on a modular approach within the framework of Principal Component Analysis (PCA). We comment on classifier performance and computational implications.;The study of the impact of image quality in face classifiers. We report on an extensive investigation aimed at revealing the relationship between classifiers performance vis-a-vis anticipated levels of image resolutions. To our knowledge this is the first investigation that quantifies the behavior of linear and non-linear face classifiers with respect to image quality. Useful design recommendations are drawn from this investigation.;The assessment of aggregation of classifiers based on image transformations. This is the first investigation putting together an assessment of numerous face classifiers and combined experts in the presence of image transformations. This is, considering both - linear and non-linear methods for constructing feature spaces. Descriptors are constructed from using contrast enhancement and edge detection. Useful findings are discussed.;The exploration of an evolutionary approach towards improving classifier performance. In this investigation we reveal in quantitative terms the importance of the features in a given face space. We present evidence of improvements in classification.;We offer useful design guidelines and recommendations for the architectures under investigation. We comment on identified advantages and drawbacks of each architecture based on the experimental findings.