AN ENHANCED PRINCIPAL COMPONENT ANALYSIS APPROACH FOR EXTRACTING FEATURES IN PALMPRINT IMAGES

Abstract

In reducing the dimensions of feature space so as to achieve better performance, minimal features that yet represent the original image maximally need to be extracted. Principal Component Analysis (PCA) which is an appearance based projection method has been widely used in this regards. Literatures have shown that PCA performs well for dimensionality reduction but instead of reducing within class variations, it increases it therefore resulting in its low performance. In contrast, Independent Component Analysis (ICA) performs preferably well in this instead since it linearly transforms an original data into completely independent components thereby reducing within class variations. While both PCA and ICA have been used for the purpose of feature extraction, no single feature extraction algorithm is exclusively flawless in all ramifications. In lieu of this, this paper introduces an Enhanced Principal Component Analysis (EPCA) approach which first extracts principal components from data and further uses the extracted components as input to an ICA algorithm. The EPCA was employed for feature extraction in a palmprint recognition system. The resultant system was validated using 900 palmprint images which were downloaded from three public palmprint databases. Recognition and verification rates arising from the palmprint recognition system showed that the EPCA performed well than PCA algorithm.