This paper discusses techniques to improve the security of gait data in biometric cryptosystems by addressing issues such as low discriminability and high data variation. The proposed methods involve utilizing Linear Discriminant Analysis and Gray Code Quantization to enhance security and reduce false acceptance rates. By preprocessing gait signals and extracting reliable features, the system aims to increase key length and reduce false rejection rates effectively.
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
ISC2017 20th Information Security Conference 22-24th November, 2017 Ho Chi Minh, Viet Nam Improving Gait Cryptosystem Security Using Gray Code Quantization and Linear Discriminant Analysis Lam Tran Lam Tran1,3 1,3, , Thang Thang Hoang Hoang2 2, , Thuc Thuc Nguyen Nguyen3 3, , Deokjai Deokjai Choi Choi1 1 1 ECE, Chonnam National University, Gwangju, South Korea dchoi@jnu.ac.kr 2 EECS, Oregon State University, Corvallis, Oregon, USA hoangmin@oregonstate.edu 3 FIT, Ho Chi Minh University Of Science, Ho Chi Minh, Vietnam {thlam,ndthuc}@fit.hcmus.edu.vn
Outline 1. 1. Introduction Introduction 2. 2. Contributions Contributions 3. 3. Proposed Proposed Methods Methods a. Solution for the Less Discriminability of Gait Data b. Method to Handle the High Variation of Gait Data 4. 4. Experiments Experiments 5. 5. Conclusion Conclusion Lam Tran et al. 2
Introduction Gait has been considered as an efficient modality for recognizing individual via human motion. Inertial sensors based gait authentication: Uses inertial inertial sensors sensors to collect gait data. Allows authenticating user implicitly Significant schemes have been proposed in literature [2]: Patten recognitions, machine learning Potential Potential risk risk: : The extracted gait templates/models are stored insecurely in mobile phone) Security and user privacy problems. Inertial sensors based gait authentication implicitly. insecurely (i.e., Lam Tran et al. 3
Introduction Fuzzy Commitment Scheme (FCS): Biometric cryptosystem framework to secure biometric templates by binding binding it it with with a a secret secret key key before before storing storing. Several studies applied FCS to the inertial-sensor based gait authentication [1]. Provide Provide elegant elegant strategy strategy to to protect protect the Drawbacks Drawbacks: : No attention to the low discriminability and high variation of gait data. Decrease the system performance (e.g., FAR, FRR) and security (e.g., key length). the stored stored gait gait templates templates. Lam Tran et al. 4
Introduction Main Main task from biometric data: Length Length: directly affects the security strength (key Discriminability Discriminability and stability stability: affects the performance Low discriminability high FAR. Low stability high FRR. task: how to extract witness witness ? (usually is a binary string) key length length) Lam Tran et al. 5
Contributions Gait signals Preprocessing & Features extraction [3] LDA Binarization Quantization Reliable bits extraction Binary string ? FCS 1. Use Linear Discriminant Analysis to solve the low discriminability of gait data. R Reduce educe FAR FAR. 2. Apply Graycode quantization to reduce the variation of gait data. Increase Increase the the key key length length and reduce reduce FRR FRR. Lam Tran et al. 6
The low discriminability Gait templates extracted from inertial-sensors based gait signals: Gait signals Noise filtering Disorientation elimination Time & domain features extraction Preprocessing & Features extraction [3] The resulted gait templates are low discriminability (the overlapping area between inter and intra class is large Ideas Ideas: : Apply LDA to enhance the discriminability. large). Lam Tran et al. 7
Fishers Linear Discriminant Analysis (LDA) LDA LDA: a data dimensional reduction technique that reserves as much as possible the discrimination discrimination information information between Given dataset ?, LDA finds a projection matrix ? to transform ? so that: Inter-class discriminability is maximized. Intra-class variation is minimized. between different different classes classes. Lam Tran et al. 8
Fishers Linear Discriminant Analysis (LDA) With a dataset ? including ? classes of ? dimensional, LDA finds projection matrix ? as: Calculate within-scatter matrix S? and between scatter matrix S?: classes, each having ?? templates x?? is template ? of class ?, x? is mean of class ?, x is mean of entire dataset. Determine the eigenvectors ??,(1 ? ?) and eigenvalues of ? is the one whose columns are have largest eigenvalues, ( W W is used to transform ? from an ? dimensional (? 1) dimensional dimensional one one. . ? ? eigenvectors ?? which ) dimensional space space to a Lam Tran et al. 9
Training data set forming Gait signals Gait signals Denote the ?th extracted gait template as: Form the training dataset: Genuine user: Preprocessing Preprocessing & Features extraction Features extraction & LDA LDA Quantization Quantization Reliable Reliable bits extraction bits extraction Other users: Binary string Binary string ? FCS FCS Lam Tran et al. 10
Problem in adopting LDA Gait signals Gait signals Problems Problems: The ?th extracted gait template: Preprocessing Preprocessing & Features extraction Features extraction & Adopting LDA: Number of class ? = 2 (genuine user and impostors) ? is formed from ? 1 = 1 eigenvector LDA LDA Quantization Quantization Reliable bits Reliable bits extraction extraction The result from LDA is a scalar Binary string Binary string ? FCS FCS Cannot quantize ? ?to a long long binary binary string string ?. Lam Tran et al. 11
LDA Training Divide the original dimensional space of gait data to ? sub apply LDA to each one. sub- -spaces spaces, then sub-space 0 sub-space 1 sub-space (? 1) LDA LDA LDA Lam Tran et al. 12
Graycode Quantizatiton The gait templates are in real inputs for FCS are binary Use quantization quantization Provide error toleration. real- -valued valued representation. The binary strings strings. Graycode Graycode: technique for designing binary numeral systems in which two two successive successive strings strings differ minimize minimize errors errors occurred occurred in in the the quantization differ in in only quantization step only one one bit step bit. Lam Tran et al. 14
Graycode Quantizatiton ?- -bit From ? , determine mean vector ? = ( normalize each component in ? to range [0,1]. Divide range value [0,1] to 2? sub-ranges, each is mapped to a unique ?-bit string following Graycode. Map each component in ? with the corresponding ?-bit string to get the quantized value. Example: 2 2- -bit bit Graycode quantization: ? 1 00 bit Graycode Graycode quantization quantization: : ? 1, , ? ?, , ? ?) , then ? 2 ? 3 ? 4 ? ? 1 00 0.75 01 0.5 11 0.25 10 0 Graycode 00 11 00 01 11 . ?? ?1 ?3 ?4 ?2 Lam Tran et al. 15
Reliable string extraction From the quantized string ??, select ? reliable ones to form the reliable string ?. ? ?, (? is codeword size of ECC). Statistical approach to calculate the reliability ?? of ??: ? = Genuine user s samples Impostors samples ? 2 ? 3 1 Intra-class variation of ? 2 is smaller than ? 3. Inter-class difference of ? 2 is larger than ? 3. ?? > ?? - 00 0.75 01 - 0.5 11 0.25 10 0 11 ?? 00 ?1 00 ?2 11 ?3 01 ?4 . ? = ?2+ ?4+ + ?? Lam Tran et al. 16
Reliable string extraction Use gait templates after LDA projection (i.e., ? ,? ) to calculate the reliability of each component. Reliability ?? of component ? is calculated as: Inter-class difference Intra-class variance Where: ? ? and ?? user s templates after LDA projection. 1 ? ?=1 ? [i,j] is value of component ? of template ? of imposters gait data. erf denotes the Gaussian error function. 2 are mean mean and variance variance of component ? of enrolled 2. 1 ? 2= ? ? [?,?] ; ? 1 ?=1 ? ?= ?? ? [?,?] ? ? Lam Tran et al. 17
Experiment and results Experiment Experiment Dataset: 38 subjects [3]. Build 38 authentication models for 38 users. LDA: Enrollment: 100 gait templates/user. Authenticating: 12 templates/trial. Graycode quantization: different values of ?. FCS: BCH BCH code code with the length of 255 and 511 bits. Different values of key length. Lam Tran et al. 18
Experiment and results Authentication Authentication results results Quantization bit number: ? = 4 Lam Tran et al. 19
Experiment and results The The impact impact of of LDA LDA projection projection Lam Tran et al. 21
Experiment and results The The impact impact of of Graycode Graycode in in quantization quantization Lam Tran et al. 22
Conclusions Addressed the low data to improve both the performance and security of gait cryptosystem. low discriminability discriminability and high high variation variation of gait Drawback Drawback: : Analyzed in laboratory. FRR is high. Future Future works works: : Experiment in real condition. Overhead time. Power consuming. Lam Tran et al. 23
References 1. T. Hoang, D. Choi, and T. Nguyen. Gait authentication on mobile phone using biometric cryptosystem and fuzzy commitment scheme. International Journal of Information Security, 14(6):549 560, 2015. S. Sprager and M. B. Juric. Inertial sensor-based gait recognition: A review. Sensors, 15(9):22089 22127, 2015. T. Hoang, D. Choi, and T. Nguyen. On the instability of sensor orientation in gait verification on mobile phone. In 12th International Conference on Security and Cryptography (SECRYPT), volume 4, pages 148 159. IEEE, 2015. 2. 3. Lam Tran et al. 24
