2D Kaiser filters with selective frequencies, selective orientations as well as changeable channels were constructed to extract iris features. The features were divided into several parts based on their amplitudes, and analysis of all the parts show that there is an ‘effective amplitude threshold’ (L) hiding in the iris features. The features with amplitudes bigger than L can achieve effective iris recognition, while those with amplitudes smaller than L are uncorrelated noises. By setting the noise features as “invalid codes”, we optimized Hamming distance and improved the correct matching rate of same pairs of irises. Results show that compared with Gabor method, the optimization method improves the right recognition rate from 98.6% up to 99.9 % and has a null fault acceptance rate with lower fault rejection rate.