Face Recognition Using Holistic Features and Within Class Scatter-Based PCA

Abstract

The Principle Component Analysis (PCA) and its
variations are the most popular approach for features clustering,
which is mostly implemented for face recognition. The optimum
projection matrix of the PCA is typically obtained by eigenanalysis
of global covariance matrix. However, the projection data
using the PCA are lack of discriminatory power. This problem is
caused by removing the null space of data scatter that contains
much discriminant information. To solve this problem, we present
alternative strategy to the PCA called alternative PCA, which
obtains the optimum projection matrix from within class scatter
instead of global covariance matrix. This algorithm not only
provides better features clustering than that of common PCA
(CPCA) but also can overcome the retraining problem of the
CPCA. In this paper, this algorithm is applied for face recognition
with the holistic features of face image, which has compact size
and powerful energy compactness as dimensional reduction of
the raw face image. From the experimental results, the proposed
method provides better performance for both recognition rate
and accuracy parameters than those of CPCA and its variations
when the tests were carried out using data from several databases
such as ITS-LAB., INDIA, ORL, and FERET.