Cross spectral iris recognition for surveillance based applications

Human iris has been explored as one of the most promising biometric traits since last many years. This paper presents a new ingenious feature extraction approach which is based on the texture variations of the iris template. A 2D Gabor filter bank is first employed to reveal the iris texture at different scales and orientations. Each filtered iris template is then partitioned into smaller sub-blocks. Contemplating the iris texture variations at micro-levels, two-level template partitioning is employed here. Difference of Variance (DoV) of corresponding first and second level sub-blocks, from each filtered image, then forms the feature set of the iris. Performance of the proposed iris recognition scheme is first tested with the benchmark IITD iris database to find the optimal window size of the filter bank. Thereafter, to prove the efficacy of the proposed approach in surveillance based applications, cross-spectral iris matching experiments (i.e. visible wavelength (VW) to near infrared (NIR) matching) are performed using PolyU cross-spectral database. Experiments show that the proposed approach achieves outperforming results for both IITD and PolyU databases.