Robust Iris Feature Extraction and Matching

An iris coding method based on zero crossings of Discrete Cosine Transform (DCT) coefficients between rectangular patches from normalized iris images is shown to provide excellent matching performance at low complexity. The method is applied to two sets of normalized iris images; 2156 images of 308 eyes from the CASIA database and 2955 images of 150 eyes from the Bath database. A product-of-sum approach to Hamming distance calculation is taken and 100% Correct Recognition Rate (CRR) achieved for identification. For verification, a variable threshold is applied to the distance metric and the False Acceptance and False Rejection Rates (FAR, FRR) recorded. The method achieves perfect Receiver Operating Characteristics (ROC), i.e. no false accepts or rejects are registered. A new metric for evaluating practical system performance is proposed and the theoretical equal error rate (EER) estimated from the Hamming Distance distributions is found to be as low as 2.59 times 10^-4.