Periocular recognition in cross-spectral scenario

Abstract

Periocular recognition has been an active area of research in the past few years due to its potential use in practical security applications. In spite of the advancements made in this area, the cross-spectral matching of visible (VIS) and near-infrared (NIR) periocular images remains a challenge. In this work, we propose a method based on illumination normalization of VIS and NIR periocular images. Specifically, the approach involves normalizing the images using the difference of Gaussian (DoG) filtering, followed by the computation of various texture based and shape based descriptors.These features include local binary patterns (LBP), histogram of oriented gradients (HOG), Gabor filter based features and local phase quantisation (LPQ) based features. Finally, the feature vectors corresponding to the query and the enrolled image are compared using the cosine similarity metric (COS) to generate a matching score. Both verification and identification experiments are performed on three publicly available benchmark databases of cross-spectral periocular images, which include IIIT Delhi multi-spectral periocular (IMP) database, PolyU cross-spectral iris database and cross-eyed periocular database. We have also investigated the performance of other existing illumination normalization methods on these databases. Our approach yields significant improvement in performance (verification accuracy) over the existing approach. We have also developed an in-house database of cross-spectral periocular images. This database includes VIS and NIR images of the left and right periocular images of 201 subjects. The performance of the proposed approach is also evaluated on this database for both verification and identification scenarios.