Challenges and Potential Research Areas In Biometrics

Transcription

1 Challenges and Potential Research Areas In Biometrics Defence Research and Development Canada Qinghan Xiao and Karim Dahel Defence R&D Canada - Ottawa October 18, 2004 Recherche et développement pour la défense Canada Canada

2 Outline Spoofing & Anti-Spoofing Liveness detection Anti-spoofing Face Recognition The most reliable passive biometric technology Problems with 2D face recognition Advantages and challenges of 3D face recognition Multimodal Biometrics Multiple biometrics Multi-modal biometrics Privacy Issue

3 Fingerprint Sensor and Spoofing Popular Fingerprint Sensors Optical sensor Capacitive sensor E-field sensor Ultrasound sensor Spoofing and Anti-spoofing

4 Optical Sensors Optical sensor is the oldest fingerprint-scanning device that consists of a source of light on one side and a camera with optical lens, on the other side to sense the transmitted light

5 The Advantages of Using CMOS Imagers Reduced Power Dissipation not only extends battery life but also lower operating temperatures Flexible Access allow to random access to a single pixel or a region of interest Increased Speed in-pixel amplifier and on-chip A/D converter produce faster frame rates Easy Integration on-chip integrated circuitry enables including processing and seeing functions in a signal package to act as a smart camera Lower Cost can be fabricated at facilities that have higher volume and need less support circuitry to operate

6 Capacitive Sensors The capacitive sensors capture the fingerprint pattern based on the measurement of the variation in capacity between skin and the sensor because there is a capacitive difference between skin-sensor and air-sensor.

7 Electric-Field (El-Field ) Sensor Electric field (e-field) sensor is built of antenna array from a new electronic fingerprint technology developed by AuthenTec. Since it examines at the subsurface of the skin, skin surface issues, such as dry, worn, or dirty skin, cannot affect the image quality.

8 Ultrasound Sensor Ultrasonic fingerprint scanner sends a short acoustic pulse toward a fingertip and measures the return echo signal. Based on the echo signal, a fingerprint image can be captured because there exists a difference in acoustic impedance between the skin and air.

9 Spoofing Problem Fingerprint sensor can be attacked by recovering latent fingerprint from sensor window, using residual prints or gelatine or silicon rubber to make fake fingers to fool the sensor

10 Fingerprint Capture Fingerprint can settle on the clean and smooth surface The problem is to make it visible Black toner is cheap and can be used to get the fingerprint It is there

11 Gummy Finger Put the plastic into hot water to soften it Press live finger against it The mold Pour the liquid into the mold Put it into a refrigerator to cool The gummy finger

12 Anti-spoofing Techniques Temperature Heartbeat Skin Resistance Detection of Finger Bone LightPrint

13 Face Recognition Problems with 2-D Facial Recognition Illumination Pose 3-D Facial Recognition 2D, 2.5D and 3D facial images Major research areas Multi-modal face recognition

14 Illumination

15 Poses

16 2D Image Normal Intensity Image of The Face

17 2.5D Image A Face Surface Representation

18 3D Image Whole Head

19 Research Areas in 3D Face Recognition Sensor Technology Increased depth of field Increased spatial resolution Reduced acquisition time New Algorithms Explore principal component analysis (PCA) Use depth information provided by 2.5D range images Investigate multi-modal face recognition

20 Multiple and Multi-modal Biometrics Multiple Biometrics Multi-input Multi-sensor Multi-algorithm Multi-instance Multi-Modal Biometrics Use more than one biometric modalities Improve the flexibility Increase the accuracy

21 Multi-modal Biometrics Fusion Templates Biometric Sensor I Feature Extractor & Match I Decision I Templates Sensor Fusion Match Decision Confidence Fusion Decision Decision Fusion Accept Reject Accept Reject Accept Reject Biometric Sensor II Feature Extractor & Match II Decision II Templates

22 Independent Or Dependent Independent/ Uncorrelated Biometric Modality I Biometric Modality II Independent/ Uncorrelated Templates Matching Algorithm I Matching Algorithm II Templates Independent/ Uncorrelated Matching Score I Matching Score II Independent/ Uncorrelated Dependent/ Correlated Dependent/ Correlated F u s i o n

23 Privacy Issues Consider essential and inessential privacy elements Build in protections for the privacy of the information in the system Owner should have control of data Minimize the possibility of misuse or abuse of biometric technology Biometric samples should never be exposed to unauthorized entities Templates and credentials should be usable only by authorized application

24 Summary on Major Challenge and Research Areas Anti-Spoofing Accuracy Privacy Data Aging Problem Standards Testing and evaluation Template exchange

25