Two-Dimensional Face Recognition Methods Comparing with a Riemannian Analysis of Iso-Geodesic Curves

摘要

In this paper, the authors performed a comparative study of two-dimensional face recognition methods. This study was based on existing methods (PCA, LDA, 2DPCA, 2DLDA, SVM...) and 2D face surface analysis using a Riemannian geometry. The last system uses the representation of the image at gray level as a 2D surface in a 3D space where the third coordinate represent the intensity values of the pixels. The authors' approach is to represent the human face as a collection of closed curves, called facial curves, and apply tools from the analysis of the shape of curves using the Riemannian geometry. Their application has been tested on two well-known databases of face images ORL and YaleB. ORL data base was used to evaluate the performance of their method when the pose and sample size are varied, and the database YaleB was used to examine the performance of the system when the facial expressions and lighting are varied.