Spatial approach to pose variations in face verification

Abstract

Spatial dimension reduction methods called Two Dimensional PCA and Two Dimensional LDA have recently been presented. These variations of traditional PCA and LDA consider images as 2D matrices instead of 1D vectors. The robustness to pose variations of these advances at verification tasks, using SVM as classification algorithm, is here shown. The new methods endowed with a classification strategy of SVMs, seriously improve, specially for pose variations, the results achieved by the traditional classification of PCA and SVM. © Springer-Verlag Berlin Heidelberg 2005.