E.A. Pavelyeva, A.S. Krylov, O.S. Ushmaev

Local iris recognition information technology using Hermite transform has been developed. The method is based on sign analysis of convolutions of iris intensity function with Hermite functions Most informative indexes (m,n) of Hermite functions and the identification criterion were justified. The developed information technology includes iris localization and normalization modules, detection of iris areas non-occluded by eyelids and eyelashes and iris rotation angle calculation. At the comparison (identification) stage binary matrixes for each index (m,n) of the fixed set are compared. As an image comparison metrics we use Hamming distance (Ham(L)) between corresponding downscaled image matrixes. Hamming distance counts the number of nonzero values in the difference of the matrixes. At first we sort the images from the database using value Ham(L1,0)+ Ham(L2,0) as the distance to the input image. Then we do the same procedure using Ham(L1,0)+ Ham(L2,1) value. If the nearest person from the database is the same for both distances the verification is positive, else the algorithm asks the user to make a photo of eye once again. A justification of the chosen index set has been performed. To make the method more robust to eye rotations we also performed limited pixel cyclical shifts for each identification procedure. As the algorithm result we find the best matching iris owner from the database. The proposed algorithm has been tested with CASIA-IrisV3 database and showed good results. It is comparable with methods currently used in practice: METHOD FAR (%) FRR (%) DATABASE Proposed 0 0.82 CASIA-IrisV3 Proposed with post-rejection 0 0.18 CASIA-IrisV3 Tan 0.001 1.13 CASIA V1.0 Wildes 0.01 6.5 Romero-Ramirez 0 9.71 CASIA V1.0 Daugman 0 0.12 NIST (ICE-1)