Adaptive noise reduction and code matching for IRIS pattern recognition system

Abstract

Among all biometric modalities, iris is becoming more popular due to its high performance in recognizing or verifying individuals. Iris recognition has been used in numerous fields such as authentications at prisons, airports, banks and healthcare. Although iris recognition system has high accuracy with very low false acceptance rate, the system performance can still be affected by noise. Very low intensity value of eyelash pixels or high intensity values of eyelids and light reflection pixels cause inappropriate threshold values, and therefore, degrade the accuracy of system. To reduce the effects of noise and improve the accuracy of an iris recognition system, a robust algorithm consisting of two main components is proposed. First, an Adaptive Fuzzy Switching Noise Reduction (AFSNR) filter is proposed. This filter is able to reduce the effects of noise with different densities by employing fuzzy switching between adaptive median filter and filling method. Next, an Adaptive Weighted Shifting Hamming Distance (AWSHD) is proposed which improves the performance of iris code matching stage and level of decidability of the system. As a result, the proposed AFSNR filter with its adaptive window size successfully reduces the effects ofdifferent types of noise with different densities. By applying the proposed AWSHD, the distance corresponding to a genuine user is reduced, while the distance for impostors is increased. Consequently, the genuine user is more likely to be authenticated and the impostor is more likely to be rejected. Experimental results show that the proposed algorithm with genuine acceptance rate (GAR) of 99.98% and is accurate to enhance the performance of the iris recognition system.