2021/09/02 by Aythami Morales, Morales, Aythami, Julián Fiérrez +7 · 1 citation
Computer Science · #User Authentication and Security Systems #Hand Gesture Recognition Systems #Biometric Identification and Security
paper · pdf · doi:10.48550/arxiv.2109.00938
This work presents a new deep learning approach for keystroke biometrics\nbased on a novel Distance Metric Learning method (DML). DML maps input data\ninto a learned representation space that reveals a "semantic" structure based\non distances. In this work, we propose a novel DML method specifically designed\nto address the challenges associated to free-text keystroke identification\nwhere the classes used in learning and inference are disjoint. The proposed\nSetMargin Loss (SM-L) extends traditional DML approaches with a learning\nprocess guided by pairs of sets instead of pairs of samples, as done\ntraditionally. The proposed learning strategy allows to enlarge inter-class\ndistances while maintaining the intra-class structure of keystroke dynamics. We\nanalyze the resulting representation space using the mathematical problem known\nas Circle Packing, which provides neighbourhood structures with a theoretical\nmaximum inter-class distance. We finally prove experimentally the effectiveness\nof the proposed approach on a challenging task: keystroke biometric\nidentification over a large set of 78,000 subjects. Our method achieves\nstate-of-the-art accuracy on a comparison performed with the best existing\napproaches.\n