On Matching Cross-Spectral Periocular Images for Accurate Biometrics Identification

Periocular recognition has gained significant importance with the increasing use of surgical masks to safeguard against environmental pollution or for improving accuracy of iris recognition. This paper proposes a new framework for accurately matching cross-spectral periocular images using Markov random fields (MRF) and three patch local binary patterns (TPLBP). We study the problem of cross spectral periocular recognition from a new perspective and our study indicates that such recognition can be considerably improved if we can preserve pixel correspondences among two matched images. The matching accuracy for cross-spectral periocular matching can be further improved by incorporating real-valued features that can be simultaneously recovered from pixels in the iris regions. We present experimental results from IIITD IMP database and PolyU database. Our experimental results validate the usefulness of this approach and achieve state-of-the-art performance for accurate cross-spectral periocular recognition.