29/01/2018 Signature Biometrics Prof. Julian FIERREZ Universidad Autonoma de Madrid - SPAIN http://atvs.ii.uam.es/fierrez Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 1 / 65 Funding Acknowledgements Public Private Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 2 / 65 1
29/01/2018 Index Introduction System Model: Pre-processing, Features, Similarity Performance Evaluation: Databases and Benchmarks Signature Aging and Template Update A Note on Tech Transfers to Industry Mobile Signature: Graphical Passwords and Swipe Biometrics Recent Advances: Signature Generation and Template Protection The Future of Behavioral Biometrics Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 3 / 65 Introduction Signature is one of the most socially accepted biometric traits, it has been used for centuries to validate legal and commercial documents and transactions Automatic signature recognition has some general challenges : Large intra-user variability ( behavioral biometric , inter-session ) Difficult to model, large amount of training data (usually scarce) Small inter-user variability (in case of forgeries) The skill level of actual forgeries is unpredictable Signatures from the same user Skilled Forgery High variability Low variability Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 4 / 65 2
29/01/2018 Introduction • High deployment of multiple electronic devices • Signatures can be easily captured by means of multiple devices • High deployment in banking and commercial sectors Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 5 / 65 Biometric Market by Modality • Decreasing (in Relative Importance ): Fingerprint , from 48% to 15% (31% w AFIS) • Growing: Iris from 9% to 16% and Face from 12% to 15% • Huge grow: Speech from 6% to 13% and Signature , from 2% to 10% Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 6 / 65 3
29/01/2018 Behavioral Biometrics • Human activity patterns are clearly stablished from childhood • As patterns, they are stable and reproducible, though subject to variability • Neuromotor coordination of gestures and movements • Continuous identity monitoring possible • User is an active part of the play • Multilevel strategy: from dynamic trajectories to expressions, context, habits, stylometry, experiences • Not fixed patterns but changing and adapting ones Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 7 / 65 Active Authentication by DARPA Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 8 / 65 4
29/01/2018 Signature as Behavioral Pattern • Human interaction permits transparent authentication • Make use of existing input channels, no added specific sensors: – Handwritting (tablets and pads) – Mouse dynamics • Other sources of variability (sensor, session) included into behavior pattern modelling / compensation • Fully revocable patterns • Incorporates soft biometrics (gender, handedness, language, …) • Easy of use, high user acceptance Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 9 / 65 Signature Recognition Azimuth (0°-359°) Altitude (0°-90°) On-line / Dynamic 0° 270° 180° 90° Off-line / Static J. Fierrez, J. Ortega-Garcia, et al., "HMM-based on-line signature verification: feature extraction and signature modeling", Pattern Recognition Letters , Vol. 28, n. 16, Dec. 2007. J. Fierrez, and J. Ortega- Garcia, “On - Line Signature Verification”, Chapter 10 in Handbook of Biometrics , A.K. Jain, A. Ross and P. Flynn (eds.), Springer, pp. 189-209, 2008. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 10 / 65 5
29/01/2018 On-line Signature Verification: Overview Feature-based (Global Features) Distance-based classifiers • Mahalanobis • Euclidean [Nelson et al., 1994] Statistical/other classifiers • Gaussian Mixture Models (GMM) • Parzen Windows Dynamic 4000 signature x 2000 0 0 50 100 150 200 250 300 350 400 2000 matching y 1000 Function-based (Local Features) 0 0 50 100 150 200 250 300 350 400 1000 z 500 0 Time-Sequence matching 0 50 100 150 200 250 300 350 400 1400 azimuth 1200 techniques 1000 0 50 100 150 200 250 300 350 400 600 altitude 500 • Hidden Markov Models (HMM) 400 0 50 100 150 200 250 300 350 400 sample index [Dolfing et al., 1998] • Gaussian Mixture Models J. Fierrez and J. Ortega-Garcia, "On-line signature verification", (GMM) [Richiardi et al., 2005] A.K. Jain et al. (Eds), Handbook of Biometrics , 2008. • Dynamic Time Warping (DTW) [Sato and Kogure, 1982] M. Martinez-Diaz and J. Fierrez, "Signature Databases and Evaluation", Stan Z. Li and Anil K. Jain (Eds.), Encyclopedia of Biometrics , Springer, pp. 1367-1375, 2015. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 11 / 65 On-line Signature Verification: System Model 1. Data Acquisition & Pre-Processing 2. Feature Extraction 3. Similarity Computation (Matching) J. Fierrez and J. Ortega-Garcia, "On-line signature verification", A.K. Jain et al. (Eds), Handbook of Biometrics , 2008. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 12 / 65 6
29/01/2018 Signature Acquisition: Input Data Time resolution: 100-200 samples/sec Space resolution: 1000 pixels/inch resolution Measured: - x,y coordinates of the signature trajectory • on pen down – time stamp at each sample point – pressure at each point – pen inclination angles at each point • altitude (0-90) • azimuth (0-359) – ... Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 13 / 65 Signature Pre-Processing Reduce sensor interoperability issues due to diverse devices and writing tools (stylus/finger) 80 60 40 Pre- - Size normalization and centering 20 processing 0 - Pressure normalization -20 -40 - Resampling -60 -80 -80 -60 -40 -20 0 20 40 60 80 M. Martinez-Diaz, J. Fierrez and S. Hangai, "Signature Features", Stan Z. Li and Anil K. Jain (Eds.), Encyclopedia of Biometrics , Springer, pp. 1375-1382, 2015. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 14 / 65 7
29/01/2018 Pre-Processing: Re-Sampling M. Martinez-Diaz, J. Fierrez and S. Hangai, "Signature Features", Stan Z. Li and Anil K. Jain (Eds.), Encyclopedia of Biometrics , Springer, pp. 1375-1382, 2015. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 15 / 65 Feature Extraction 27 local feature set M. Martinez-Diaz, J. Fierrez, et al., "Mobile Signature Verification: Feature Robustness and Performance Comparison", IET Biometrics , Dec 2014. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 16 / 65 8
29/01/2018 Feature Extraction: Global Features X Y P Az Al 0 100 200 300 M. Martinez-Diaz, J. Fierrez, et al., "Mobile Signature Verification: Feature Robustness and Performance Comparison", IET Biometrics , Dec 2014. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 17 / 65 Feature Extraction: Global Features Example 2.5 0.6 Genuine Signatures (All) Skilled Forgeries (All) 0.5 2 Genuine Signatures (Shown) Skilled Forgery (Shown) 0.4 1.5 0.3 Global Feature 2 Global Feature 4 0.2 1 0.1 0.5 0 -0.1 0 -0.2 -0.5 -0.3 -1 -0.4 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 Global Feature 1 Global Feature 3 Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 18 / 65 9
29/01/2018 Feature Extraction: Global Features Example 25 0.7 Genuine signatures from all users Genuine signatures from all users Specific user signatures Specific user signatures Skilled forgeries 0.6 Skilled forgeries Average pen speed (Feature num. 26) Number of pen-ups (Feature num. 2) 20 0.5 15 0.4 X 0.3 10 Y 0.2 P 5 Az 0.1 Al 0 0 0 100 200 300 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0 50 100 150 200 250 300 350 400 450 Signature duration (Feature num. 1) M. Martinez-Diaz, J. Fierrez, et al., "Mobile Signature Verification: Feature Robustness and Performance Comparison", IET Biometrics , Dec 2014. Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 19 / 65 Global Features: Performance (on MCYT DB) 5 training signatures 20 training signatures 12 12 Global (Parzen) 11 11 Local (HMM) 10 10 9 9 8 8 EER (%) EER (%) 7 7 6 6 SKILLED 5 5 4 4 3 3 2 2 1 1 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 5 5 Global (Parzen) 4.5 4.5 Local (HMM) 4 4 3.5 3.5 EER (%) EER (%) 3 3 2.5 2.5 RANDOM 2 2 1.5 1.5 1 1 0.5 0.5 0 0 10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100 Julian Fierrez – Winter School on Biometrics, Shenzhen, CHINA – Jan. 2018 – Slide 20 / 65 10
Recommend
More recommend
