An improved algorithm for iris feature extraction

Abstract

2D Log-Gabor will produce a crescent-shaped waveform in frequency domain if its angular bandwidth approximates 2π. The grid-like shape will then appear in iris feature images, which reduces the validity of the features. To solve the problem, this paper proposes a new algorithm for iris feature extraction. The algorithm uses the odd-symmetric 2D Log-Gabor filter to analyze the phase and amplitude of iris texture of different frequencies and orientations, and uses feature fusion to eliminate redundant feature. It successfully prevents the crescent-shaped spectrum and grid-like patterns. The experimental results show that the equal error rate (EER) of the proposed algorithm can be 11% and 38% lower than the 2D Gabor filter and Log-Gabor composite filter respectively on average. It clearly justifies the strong robustness of the proposed algorithm against noise and interferences.