Iris Recognition under Non-ideal Imaging Conditions and CCD Noise

Iris recognition is the most reliable and dependable biometric system as the features of human eye are invariant and distinctive for every individual. Present iris recognition algorithms are tested using the bench mark databases which are assumed to be almost ideal except for eyelid and eyelash occlusions and rotational inconsistencies. It has been discussed elaborately by Daugman in [3] that, non-ideal imaging conditions affect the ”authentics” distribution in the decision environment graph. Getting motivation from this observation, all possible non-ideal imaging conditions and Charge Coupled Device (CCD) noise are simulated and applied on the available databases. Legendre moments, introduced by Teague can achieve translation and scale invariance and also, close to zero value of redundancy measure, so that the moments correspond to distinct and autonomous features of the image. In the proposed method it is proved that, they can also work very well on noise affected features when trained and tested using SVMs. The performance of the Exact Legendre Moments (ELM) on UBIRIS and CASIA datasets proves to be very good with Correct Recognition Rate (CRR) = 99.6% under non-ideal imaging conditions and CCD noise.