Representational Oriented Component Analysis (ROCA) for Face Recognition with One Sample Image per Training Class - Robotics Institute Carnegie Mellon University

Representational Oriented Component Analysis (ROCA) for Face Recognition with One Sample Image per Training Class

Fernando De la Torre Frade, Ralph Gross, Simon Baker, and Vijaya Kumar
Conference Paper, Proceedings of (CVPR) Computer Vision and Pattern Recognition, Vol. 2, pp. 266 - 273, June, 2005

Abstract

Subspace methods such as PCA, LDA, ICA have become a standard tool to perform visual learning and recognition. In this paper we propose Representational Oriented Component Analysis (ROCA), an extension of OCA, to perform face recognition when just one sample per training class is available. Several novelties are introduced in order to improve generalization and efficiency: 1) Combining several OCA classifiers based on different image representations of the unique training sample is shown to greatly improve the recognition performance. 2) To improve generalization and to account for small misregistration effect, a learned subspace is added to constrain the OCA solution. 3) A stable/efficient generalized eigenvector algorithm that solves the small size sample problem and avoids overfitting. Preliminary experiments in the FRGC Ver 1.0 dataset (http://www.bee-biometrics.org/) show that ROCA outperforms existing linear techniques (PCA,OCA) and some commercial systems.

BibTeX

@conference{Frade-2005-9188,
author = {Fernando De la Torre Frade and Ralph Gross and Simon Baker and Vijaya Kumar},
title = {Representational Oriented Component Analysis (ROCA) for Face Recognition with One Sample Image per Training Class},
booktitle = {Proceedings of (CVPR) Computer Vision and Pattern Recognition},
year = {2005},
month = {June},
volume = {2},
pages = {266 - 273},
keywords = {Face Recognition, one training sample, Oriented Component Analysis,},
}