Human identification from at-a-distance face images using sparse representation of local iris features

Automated human identification at-a-distance, using completely automated iris segmentation, is highly challenging and has wide range of civilian and forensics applications. Iris images acquired at-a-distance using visible and infrared imaging are often noisy and suffer from divergent spectral changes largely resulting from scattering, albedo and spectral absorbance selectivity. Therefore further research efforts are required to develop feature extraction techniques which are more tolerant to illumination changes and noise. This paper develops a new approach for the automated recognition from such distantly acquired iris images using sparse representation of local Radon transform (LRT) based orientation features. We model the iris representation problem as sparse coding solution based on computationally efficient LRT dictionary which is solved by widely studied convex optimization approach/strategy. The iris recognition and verification performance for the distantly acquired iris images are also evaluated using baseline 1-D log-Gabor filter and monogenic log-Gabor filter based approach. The experimental results are reported on the publically available UBIRIS V2, FRGC and CASIAV4-distance databases. The achieved experimental results on at-a-distance databases are highly promising and confirm the usefulness of the approach.