Time-Frequency Features of Laplacian Decomposition

Smartphones are gaining popularity as a biometric authentication device for many applications like banking and e-commerce. While multiple smartphone manufacturers providing phones with new features, consumers venture to try new phones or use multiple devices to sign into secure applications leading to a situation, in which they have to deal with data emerging from various smartphones along with manifold capture conditions. Recognition processes involving such cross-smartphone data leads to degraded biometric performance. In this work, we propose a novel technique to extract texture features from the periocular region by decomposing the images into Laplacian pyramids of various scales and obtain frequency responses in different orientations. Further, we propose to encode the features sparsely as a means to increase the cross-smartphone authentication using periocular region. From the extensive set of experiments conducted on a publicly available smartphone periocular database, we demonstrate the improvement using the proposed feature encoding and comparison for authentication scenarios. An average gain in Equal Error Rate of around 10 % is achieved while the best gain of 16 % is obtained for various comparisons as compared to previously reported verification scores. The obtained gain in performance indicates the applicability of the proposed feature extraction technique for real-life authentication scenarios employing data from different smartphones in the visible spectrum.