Dynamic scale selected Laplacian decomposed frequency response for cross-smartphone periocular verification in visible spectrum

Secure applications like banking and e-commerce allow users to access applications from multiple devices. Growing popularity of biometrics for such applications results in users providing biometrics data from multiple devices of different make and manufacture (smartphones or tablets). The performance of biometric systems degrade generally when the enrollment data and probe data originate from different devices such as different make of smartphones due to different image characteristics, varying capture conditions with respect to illumination and user interaction. In this work, we investigate the performance of cross-smartphone periocular verification and propose a new technique to improve the verification performance for images originating from different smartphones. The proposed scheme is based on employing spatial and frequency components obtained after the Laplacian decomposition of the image coupled with dynamic scale selection. The proposed technique is evaluated on the publicly available visible spectrum iris (periocular) database - MICHE I dataset. Equal Error Rate (EER) is reduced by an average of 14% and 12% for iPhone and Samsung data respectively as compared to previously published results. The improved performance indicates the contribution of scale selection for such cross-smartphone and cross-scenario periocular verification.