A novel phase-intensive local pattern for periocular recognition under visible spectrum

The article proposes a novel multi-scale local feature based on the periocular recognition technique which is capable of extracting high-dimensional subtle features existent in the iris region as well as low-dimensional gross features in the periphery skin region of the iris. A set of filter banks of different scales is employed to exploit the phase-intensive patterns in visible spectrum periocular image of a subject captured from a distance in partial non-cooperative scenario. The proposed technique is verified with experiments on near-infrared illumination databases like BATH and CASIA-IrisV3-Lamp. Experiments have been further extended to images from visible spectrum ocular databases like UBIRISv2 and low-resolution eye regions extracted from FERETv4 face database to establish that the proposed feature performs comparably better than existing local features. To find the robustness of the proposed approach, the low resolution visible spectrum images of mentioned databases are converted to grayscale images. The proposed approach yields unique patterns from these grayscale images. The ability to find coarse-to-fine features in multi-scale and different phases is accountable for the improved robustness of the proposed approach.