Orthogonal moments for efficient feature extraction from line structure based biometric images

A striking feature in most widely used biometric images such as fingerprint and palmprint are certain prominent line structures. These structures are in the form of arches, whorls and loops in fingerprints while line segments are in palmprint. This paper makes use of orthogonal moments, namely Legendre, Pseudo-Zernike and Chebyshev moments, to extract features from this type of biometric images. These moments are widely used as shape descriptors. Bayesian Belief Net (BBN) is used to classify the moment based features. Experimental results reveal that features extracted from these line structure based images with the help of orthogonal moments are found to be very accurate and can be used for individual identification. It also analyzes the performance of the multimodal biometric systems by making feature level fusion of moments from fingerprints and palmprints.