2. Contents
1. Defining the Program Requirements PAUL
2. Facial Recognition & Biometric Encryption KLAUS
3. The Human Side of Detection PAUL
4. Future Direction KLAUS
RG Overview Page 2
4. Defining Program Requirements
Self-Exclusion
Option to take a break from slot/casino gambling
Self-help tool for players who are working to control behaviour
OLG Role:
Provide clear information, implications for entering
Effectively deliver systems, policies, procedures
Stop direct marketing
Provide referrals as a “gateway” to a system of
community service that are “individually tailored”
Applies to slots/casino sites in Ontario
Does NOT apply to lottery, bingo, horse racing
RG Overview Page 4
5. Defining Program Requirements
What Self-Exclusion is Not
Determination/judgement about a gambling problem
A policing program
A way to prevent people from gambling
Dr. Howard J. Shaffer
Harvard Medical “… responsibility for self-exclusion and
School – ultimately gambling remains with the
Division on patron… Even the name, self-exclusion,
Addictions should serve to remind patrons, policy
makers and industry observers that the
responsibility for the behaviour of the
gamblers who enroll in self-exclusion
programs remains with them.”
RG Overview Page 5
6. Defining Program Requirements
Why Attempt to Detect Self-Excluders at
All?
Deterrent to Breaching NOT Policing
If a self-excluded person is detected, s/he will be escorted
from site, and can be trespassed
Support by operator includes creating disincentives to
breaching
FR is not the answer on its own… it is on part of
an overall perimeter of support for Self-Excluders
RG Overview Page 6
7. Defining Program Requirements
Context for Facial Recognition
Objectives: To support players, evolve practices, build
corporate reputation
OLG decisions must consider:
Vulnerable Player
Most Self-Excluders have significant problems
Segment
Program Standards International dialogue on best practices
Brand Integrity OLG is highly scrutinized
RG Overview Page 7
8. Defining Program Requirements
Program Priorities
Decision to implement FR required the following criteria:
• Sufficient “true hit” rate
SYSTEM • Acceptable “false positive” rate
PERFORMANCE • Defensible cost
• “Privacy by design” approach
RESPECT for • Protection of images/data to exceed industry standards
PRIVACY • Images of non-self-excluders had to be deleted
• Security officers use terminals at podium
EASE of • System allows officers to review images that
OPERATION appear with a “hit”, in order to “make call”
• Operate seamlessly with surveillance systems
RG Overview Page 8
9. Defining Program Requirements
Partners in Facial Recognition
Information
AGCO
Privacy
Regulator
Commissioner
University of iView
Toronto Systems
RG Overview Page 9
11. Self Exclusion Technology Timeline
Overall Approach
Online and Privacy FR+BE Build Rollout
FR+BE
centralized requirements FR+BE solution production FR+BE
proven
SE system finalized tuned confirmed FR+BE technology
viable
(no FR) by IPC system at OLG
»live April »minor reduction »meetings with »80% to 90% CIR »proposal »rollout results are
2009 in CIR IPC staff for OLG volunteer vetted by iView, consistent with POC tests
»~50% reduction group UofT, IPC and
in false alarms »detecting 30 times OLG Exec
more SE than the
current process
• Measured approach to developing the system
• Privacy by Design
• Used staff control groups to measure system performance
• Lighting and pose are key to facial recognition success
• Field trial at Woodbine to validate system performance
• Rollout to all sites
RG Overview Page 11
12. Privacy by by Design:
Privacy Design Privacy + Security
We are discarding all
captured images except
correctly recognized alerts
HASH
FR PI
IMAGE
NAME
TMPL
ADDR
BE
RG Overview Page 12
13. Face Recognition Performance
Control Group Results
Correct Identification Rate
30% Lighting
Improvements
49% Improvements 80%
Entrance Additional
Lighting
88% Camera
Positioning
91%
»baseline at »test at »test at »test at »test at
Casino SSM Foster Drive Woodbine Slots Woodbine Slots Woodbine Slots
Apr. 2009 Oct. 2009 Oct. 2009 Oct. 2009 Mar. 2010
Note: All tests were controlled by using volunteer OLG employees to determine the Correct Identification Rate
RG Overview Page 13
15. Human Side
Role of Security Officers
Must capture image correctly DATA
Carry out registration accurately INPUT
Potential “hits” appear on terminal REVIEW
Review and decide “HITS”
Confirm identity on gaming floor
Complete the breach
INTERCEPT
Appropriate reporting
RG Overview Page 15
16. Human Side
Duty of Care Implications?
Detecting SE who breach is a requirement of SE
program–a support to discourage return to gaming sites
Photos in binders is one way to do this, FR is another
Duty of Care/Standard of Care
RG Overview Page 16
18. Ensuring Performance
Mystery Shop program with credible independent 3rd
party
Technology and pattern reviews to augment the
technology base
Product upgrades to implement industry FR
enhancements
RG Overview Page 18
19. Rollout and other Options
Approximately 20 sites remaining
Scheduled for completion Q2 of this fiscal
Off-site registration Process
Facial recognition can be extended to other areas of the
casino – for example kiosks, non entrance locations, etc
Extend the facial recognition technology to other populations
Optimize the application for mobile platforms
RG Overview Page 19
20. FR/BE Health Check and Enhancements
Post rollout review and tuning is an ongoing task
Privacy audit to validate the system design and
implementation
Site adjustments – optimized and/or additional cameras
As detection levels fluctuate, understand why – SE program
success versus system performance problems
Analysis, analytics and trending for RG and addiction research
RG Overview Page 20
21. Additional Sources
OLG/IPC paper:
Privacy-Protective Facial Recognition: Biometric Encryption Proof of Concept
http://www.ipc.on.ca/images/Resources/pbd-olg-facial-recog.pdf
IEEE pub:
Martin, K., Lu, H., Bui, F., Plataniotis, K. N. and Hatzinakos, D.: A biometric encryption system
for the self-exclusion scenario of face recognition. IEEE Systems Journal: Special Issue on
Biometrics Systems 3(4), 440-450 (2009)
IEEE pub:
Lu, H., Martin, K., Bui, F., Plataniotis, K. N. and Hatzinakos, D.: Face recognition with
biometric encryption for privacy-enhancing self-exclusion. (2009)
IEEE pub:
Bui, F.M., Martin, K., Lu, H., Plataniotis, K.N., and Hatzinakos, D.: Fuzzy Key Binding
Strategies Based on Quantization Index Modulation (QIM) for biometric Encryption (BE)
Applications. IEEE Transactions On Information Forensics and Security 5(1), 118-132 (2010)
RG Overview Page 21