Iris and Periocular Recognition using Shape Descriptors and Local Invariant Features

Abstract

Iris is a popular biometric modality that has been deployed in uncontrolled environment for various applications like the Aadhaar project in India, in airports, banks, health and educational institutes. However, occlusion of eyelids, eyelashes and illumination variations result in degradation of biometric system recognition. Thus, another biometric modality ‘periocular’ has been proposed in the paper in complementary to ‘iris’ modality. ‘Periocular’ refers to the region surrounding eye i.e. eyelids, eyelashes, eyebrows and skin texture. A periocular database consisting of 1000 images has been prepared. The paper proposes a feature-set consisting of shape descriptors: Local Binary Pattern (LBP) and Scale-Invariant Feature Transform (SIFT) along with orthogonal moments like Zernike, Krawtchouk, Tchebichef and Dual-Hahn. The feature-set is concatenated and fed into a K-NN classifier. Experiments are performed on publicly available database: IIITD Multi-spectral periocular and self-developed PEC periocular database. Results demonstrate that Dual-Hahn moments show recognition accuracy of 97.8% for IIITD database and Tchebichef moments show an accuracy of 92.7% for PEC periocular database. The proposed method achieves superior results when compared to other methods available in literature.