Enhanced accuracy moment invariants for biometric recognition and cryptosystems

Abstract

Numerical accuracy of moment invariants is very important for reliable feature extraction in biometric recognition and cryptosystems. This paper presents a novel approach to derive accuracy enhanced moment invariants that are invariant under translation, rotation, scaling, pixel interpolation and image cropping. The proposed approach defines a cosine based central moment and adopts a windowing mechanism to enhance accuracy of moment invariants under translation, rotation, scaling, pixel interpolation and image cropping. It derives moment invariants by extending the knowledge used in Hu's and Maitra's approaches. Simulation results show that the proposed moment invariants highly accurate than Hu's and Maitra's moment invariants. © 2009 Springer Berlin Heidelberg.