Chen, Ph.D.) Visual Information Processing & CyberCommunications Lab. (VIP-CCL) 視覺資訊處理暨信息通訊實驗室.

Transcription

1 Face Recognition (Wen-Shiung Chen, Ph.D.) Visual Information Processing & CyberCommunications Lab. (VIP-CCL) 視覺資訊處理暨信息通訊實驗室 1 OUTLINE Introduction Biometric Recognition Face Recognition: Introduction Methods: Briefly Some New Applications Conclusion 2

2 (Introduction) 3 Biometrics 4

3 Introduction (personal/identity authentication) (e-authentication) (CSI; Forensics) ATM 5 Introduction Especially, after New York 911,.. 6

4 Personal Identification Objects Token-based: Something that you have Key, smart card, magnetic card, passport, USB token, etc. Knowledge-based: Something that you know Password, PIN,, etc. Biometrics-based: Something that you are Fingerprint, Face, Iris, Voiceprint,, etc. Introduction : Token-based and Knowledge-based Key License or card Password.. so on Lost Stolen Forgotten (too many, hard to memorize) Easy to memorize, easy to guess! Misplaced Good Solution: Biometrics! 8

5 (Biometric Recognition) 9 Funny, But True Boss, may I have your My passwords are 10

6 (Biometrics) Bio + metrics: The statistical measurement of biological data. Biometric Consortium definition: Automatically recognizing a person using distinguishing traits. (Biometric Recognition) Biometrics is the measurement and statistical analysis of biological data In IT, biometrics refers to technologies for measuring and analyzing human body characteristics for identity authentication purposes (Pattern Recognition) (Biometric Recognition) 12

7 Where the Biometric Traits? Head Hand 13 Introduction (biometric recognition) No need to remember passwords Unauthorized access to personal data can be prevented Fraudulent use of ATMs, credit cards can be prevented No need something to token based systems 14

8 (Physiological/Static Characteristics) biometric methods authentication based on a feature that is always present (Behavioral/Dynamic Traits) biometric methods authentication based on a certain behaviour pattern 15 Biometric Systems (Enrollment) (Recognition) (Authentication) Identification (Recognition) Verification (Authentication) Authentication Identification Verification Recognition 16

9 Enrollment Person entered into the database 17 Identification Who am I? or Who is this guy? 1-to-many mapping Example: FBI 18

10 Verification Am I who I claim to be? 1-to-1 comparison Confirms a claimed identity Claim identity using name, user ID, Example: 19 Possessions Knowledge Manual and semiautomated biometrics Characteristics Biographics Automated biometrics Physiological Bahavioral Face DNA Fingerprint Eye Hand Signature Voice Keystroke Retina Iris Typology of identification methods Hand Shape Palmprint Vein 指靜脈 20

11 Face Fingerprint Handshape/Palmprint Iris Infrared Image (IRID) Gait BIOMETRICS Retina Hand Vein Signature Voiceprint 21 Fingerprint Recognition Thermal sensor [FingerChip by ATMEL (was: Thomson CSF)] [TravelMate 740 by Compaq und Acer] Optical fingerprint sensor [Fingerprint Identification Unit FIU-001/500 by Sony] [BioMouse Plus by American Biometric Company] 22 E-Field Sensor [FingerLoc by Authentec]

12 Hand Recognition Palmprint ( ) Hand shape ( ) Hand geometry reader by Recognition Systems 23 Hand geometry reader for two finger recognition by BioMet Partners Iris Recognition Iris ( ) Feature iris (Pupil) (Iris) 1999 (ATM) Stella NCR 24

13 Retina Recognition Retina ( ) Feature retina Retinal recognition system [Icam 2001 by Eyedentify] 25 Voiceprint Recognition Voiceprint Recognition ( ) Speaker Recognition ( ) Fixed text Text dependent Text independent Conversational 26

14 Handwritten Recognition Handwritten or Signature ( ) Static (off-line) Signature Recognition Electronic pen [LCI-SmartPen] Dynamic (on-line) Signature Recognition (DSR) 27 Gait Recognition Gait ( ) 28

15 DNA Recognition DNA (non-technical) DNA DNA contains information about race, paternity, and medical conditions for certain disease Ultimate Identifier Technical Problems Not yet fully automated, not fast (not realtime) and expensive Theoretical limitation: Identical twins have the same DNA 29 Infrared Thermogram Image Infrared Facial Thermograms Identical twins have different thermograms Hand Vein Thermograms 30

16 Ear Recognition Feature vector of distances of salient point Ear geometry recognition uses the shape of the ear to perform identification An infrared image can be used to eliminate hair Might be recognized at a distance 31 Keystroke Dynamics 32

17 Face Recognition Analyzes some keypoints in human face, and relationship between them Analyzes color Face recognition system [TrueFace Engine by Miros] Face recognition system [One-to-One by Biometric Access Corporation] 33 UCSD Biometric Soda Machine Face Recognition 34

18 Video Surveillance (Airports) Face Recognition 35 New Passports Face Recognition 36

19 Access Control Hand-shape Recognition 37 Fingerprint System at Gas Stations Galp Energia SGPS SA of Fingerprint Recognition Lisbon won the technology innovation award for developing a payment system in which gasoline-station customers can settle their bills simply by pressing a thumb against a glass pad. Scanning technology identifies the thumbprint and sends the customer's identification information into Galp's backoffices system for payment authorization. THE WALL STREET JOURNAL, November 15,

20 Fingerprint System at Border Crossing Fingerprint Recognition 39 Biometrics for Personalization Fingerprint Recognition 40

21 Iris Scans to Unlock Hotel Rooms Iris Recognition 41 The Nine Zero hotel in Boston just installed a new system which uses digital photos of the irises of employees, vendors and VIP guests to admit them to certain areas, the same system used in high-security areas at airports such as New York's JFK Want to Charge It? Voiceprint Recognition Then talk to your credit card 42

22 Did You Vote? 43 (Face Recognition Technology) 44

23 Face Recognition Query: Who is this guy? Ans: Bill Clinton. Ans: Monica Lewinsky. 45 Example Face Identification Is she in the database? YES! 46

24 Face Recognition System 47 Face Recognition: Correlation 48

25 Introduction Why Face Recognition? Natural and easy to use Many potential applications, such as person identification, human-computer interaction, security systems Stages of Face Recognition Face location detection Feature extraction Facial image classification Approaches of Feature Extraction Local features Eyes, nose, mouth information Easily affected by irrelevant information Global features Extract features from a whole image 49 Introduction Advantages Most natural method: without user cooperation Non-intrusive ( ) Low cost Ability to operate covertly Disadvantages Affected by appearance/environment High false non-match rates Identical twins attack Potential for privacy abuse 50

26 Introduction Face recognition from images is a sub-area of the general object recognition problem Approaches of Feature Extraction Local features Eyes, nose, mouth information Easily affected by irrelevant information Global features Extract features from a whole image Applications law enforcement personal identification driver s licenses credit card gateway of limited access areas Some new applications on multimedia 51 etc. Enrollment Feature Database Biometric Reader Face Detection Feature Extraction Recognition Biometric Reader Pre-processing Face Detection Feature Extraction Pattern Recognition Decision 52

27 Just Face Detection? Cool! Face Detection! So strong the power of camera is! 53 Face Detection Face Detection! Camera tell us where the faces are! How smart the camera! 54 Ask the machine: Where are faces? and male or female?

28 Locating Faces in a Crowd A difficult problem! 55 Face Detection Face Detection and Localization Facial image Arbitrary image Problems Orientation of the face Causal and complex scene Illumination variation Computing Methods Histogram: vertical and horizontal Brightness distribution Edge detection (point distribution model, PDM) Integral projection on color and edge information 56

29 Feature Extraction: Approaches Methods Manually defining features --- Feature-based Automatically deriving features --- Appearance-based Feature-based Local model based on Face Geometry (geometrical features) The idea is to model a human face in terms of particular face features, such as eyes, nose, mouth, etc., and the geometry of the layout of these features Local features: Eigen-eyes or Eigen-noses, etc. Appearance-based (or Template-based) Global model based on Face Appearance or template The underlying idea is to reduce a facial image containing thousands of pixels to a handful of numbers To capture the distinctiveness of the face without being overly sensitive to noise such as lighting variations A face image is transformed into a space that is spanned by basis image functions, e.g., Fourier transform, KLT, wavelet, Eigenfaces using KLT or PCA (by Turk and Pentland, MIT) 57 Face Data Extraction Static Images and Feature-Based Methods FCPs : Facial Characteristic Points -by Kobayashi Facial Points of the frontal-view face model and the side-view face model -by Pantic 58

30 Face Data Extraction Static Images and Template-Based Methods A small model-graph (small GFK) -by Hong Dense model-graph (big GFK) Method : 1. Gabor wavelet extracted at a point of the input image 2. GFK : General Face Knowledge 3. Small GFK : find the exact face location in a facial image 4. Big GFK : localize the facial feature 5. Real-time process 59 (Eigenfaces Eigenfaces) Math. Basis: Principal Component Analysis (PCA) (Eigenfaces) 60

31 Eigenface Recognition System MIT Media Laboratory Vision and Modeling Group 61 Eigenface: : An Example A face image can be represented exactly as weighted combinations of the eigenface components e 1 e 2 e 3 e 4 W 1 W 2 W 3 W 4 Original Face image 62

32 Problems and Challenges camera camera Camera 63 Problems: Lighting Variability Indoor Outdoor 64

33 Same person or different person? 65 Same person or different person? 66

34 Same person or different person? 67 Intra-Class Variability 68

35 Inter-class similarity (Twins) 69 Temporal Variations 70

36 Applications: Issues of Recognition Traditional Application Issues Identity Authentication Recognize a person who he or she is? Facial Expression Others: Hat, Glasses, Beard, Mustache, New Emerging Application Issues Gender Classification Recognize sex: male or female? Age Estimation and Classification Recognize a person how old one is? Ethnicity (Race) Classification Recognize race or skin color Genealogy Verification Recognize two faces with similar look Father-son, mother-daughter, sibling, 71 Gender Classification Problem statement Determine the gender of a subject from facial images Female Potential applications Face Recognition Human-Computer Interaction (HCI) Challenges Race, age, facial expression, hair style, etc. Male 72

37 Can You Tell? Not easy to do, even human eyes! 73 Can You Tell? Answer: F - M - M - F - M Not a easy problem! 74

38 Component-based Classification Component-based gender classification Local model based on Face Geometry (geometrical features) 75 Classification using SVM Appearance-based gender classification Global model based on Face Templates Learning to classify pictures according to their gender (Male/Female) when only the facial features appear (almost no hair) 76

39 Results 77 An Application on Multimedia NEC Eye flavor 78

40 NEC Eye Flavor 79 NEC Eye Flavor 80

41 Conclusion Just One Word: Face Recognition is very Crucial and Valuable from both Academic and Commercial Viewpoints! 81 Thank You Very Much! 4W Beautiful Women Sweet Water Good Wine Nice Weather 82

42 Q & A 83