Abstract
Three different localized representation methods and a manifold learning approach to face recognition are compared in terms of recognition accuracy. The techniques under investigation are (a) local nonnegative matrix factorization (LNMF); (b) independent component analysis (ICA); (c) NMF with sparse constraints (NMFsc); (d) locality-preserving projections (Laplacian faces). A systematic comparative analysis is conducted in terms of distance metric used, number of selected features, and sources of variability on AR and Olivetti face databases. Results indicate that the relative ranking of the methods is highly task-dependent, and the performances vary significantly upon the distance metric used.
Highlights
Face recognition has represented for more than one decade one of the most active research areas in pattern recognition
The list is augmented by representation procedures using space-localized basis images, three of which are described in the present paper; (c) the assumption that many real-world data lying near low-dimensional nonlinear manifolds exhibiting specific structure triggered the use of a significant set of manifold learning strategies in face-oriented applications [9, 10], two of which are included in the present comparative analysis
Images AR02, AR03, and AR04 are used for testing the performances of the analyzed techniques to deal with expression variation, images AR05, AR06, and AR07 are used for illumination variability, and the rest of the images are related to occlusion, with variable illumination conditions
Summary
Face recognition has represented for more than one decade one of the most active research areas in pattern recognition. A number of recent surveys [1, 2] review modern trends in this area of research, including (a) kernel-type extensions of classical linear subspace projection methods such as kernel PCA/LDA/ICA [3,4,5,6];. The present paper focuses on a systematic comparative analysis of subspace projection methods using localized basis functions, against techniques using locality-preserving constraints. We have conducted extensive computer experiments on AR and Olivetti face databases and the techniques under investigation are (a) local nonnegative matrix factorization (LNMF) [12]; (b) independent component analysis (ICA) [13]; (c) nonnegative Matrix Factorization with sparse constraints (NMFsc) [14]; and (d) locality-preserving projections (Laplacian faces) [9]. We have taken into account a number of design issues, such as the type of distance metric, the dimension of the feature vectors to be used for actual classification, and the sources of face variability
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
