Biometric feature extraction with biometric specific shape descriptors

Abstract

Orthogonal moments, namely Legendre and pseudo-Zernike moments, are popular region-based shape descriptors which can be used to represent regions with invariant feature vectors. In this paper we propose that the accuracy of biometric verification can be improved by selecting a shape descriptor after exploring its suitability for the given shape or biometric. Fingerprint and palmprint biometrics are used for the experiment. It was observed that the inherent nature of the Legendre moments makes it more suitable for palmprint images and that of the Zernike moments makes it more suitable for the fingerprint images.