An Empirical Study of Dimensionality Reduction Methods for Biometric Recognition

Abstract

This research aims at studying the recognition accuracy and execution time that are affected by different dimensionality reduction methods applied to the biometric image data. We comparatively study the fingerprint, face images, and handwritten signature data that are pre-processed with the two statistical based dimensionality reduction methods: principal component analysis (PCA) and linear discriminant analysis (LDA). The algorithm that has been used to train and recognize the images is support vector machine with linear and polynomial kernel functions. Experimental results showed that the application of LDA dimensionality reduction method before recognizing the image patterns with a linear kernel function of SVM is more accurate and takes less time than the recognition that did not use dimensionality reduction. LDA is a suitable technique for physiological biometrics, whereas PCA is appropriate for the behavioral biometrics. We also found out that only 1% of transformed dimensions is adequate for the accurate recognition of biometric image patterns.