On the dimensionality of PCA method and color space in face recognition

Abstract

The paper presents a problem of color images recognition in the aspect of dimensionality reduction performed by means of different Principal Component Analysis variants. The aim of the experiments was to check the applicability of one- and two-dimensional PCA in color image classification together with an analysis of an employed color-space. Since, most of the works in this area are focused on grayscale images, in this paper we investigate several full-color representations, as it improves the overall recognition rate. As a comparison, typical approaches involving one-dimensional PCA and two-dimensional PCA on RGB, HSV, YIQ and YCbCr representations of images is provided. The paper describes theoretical fundamentals of the algorithm and implementation for these variants of PCA. Furthermore, the impact of the number of principal components on the recognition accuracy is investigated. The usefulness of PCA on color images is investigated on typical benchmark databases containing facial portraits. © 2013 Springer-Verlag.