Minutiae Template Interoperability Testing

1
Minutiae Template Interoperability Testing

• CEN Biometrics Focus Group5 December 2006

2
MTIT Objectives

• Work with EC to provide for testing and
  certification define criteria for
  interoperability testing
• define criteria for interoperability testing
• develop a database of fingerprint images
• develop a test bed for fingerprint minutiae
  interoperability
• incorporate an improvement step
• provide for the testing of interoperability of
  future systems

Co-funded by the European Commission within the
Sixth Framework Program (STREP) - Project no.
IST-2005-027351
3
Scientific Objectives of MTIT

• Conduct research to deliver encoding algorithms
  able to provide interoperability
• Deliver a framework and a certification process
  which allow future interoperability
• independent certification laboratory will conduct
  such tests

Co-funded by the European Commission within the
Sixth Framework Program (STREP) - Project no.
IST-2005-027351
4
List of Participants
Co-funded by the European Commission within the
Sixth Framework Program (STREP) - Project no.
IST-2005-027351
5
Conformance Testing
Automated conformance checking
GUI-based conf. checking
Conf. check I/F
DaonAnalytics
Extract, filter, score quality
Generate templates
Automatic conformance check
Template matching
Compute accuracy and graph
Finger Database
MIT Test Interface
Cogent
Motorola
NEC
Sagem
6
Algorithms to be Analysed

• Proprietary claim template, proprietary enrolment
  template
• Proprietary claim template, record format
  enrolment template
• Proprietary claim template, card format enrolment
  template
• Record format claim template, record format
  enrolment template
• Card format claim template, card format enrolment
  template

7
18 Databases for Analyses
8
Main findings 1

• Large variations in performance between different
  databases
• Difficult to define interoperability purely on
  basis of performance
• (As per MINEX) Proprietary2Fingermore accurate
  than Minutiae2fingermore accurate than
  Proprietary1Fingermore accurate than
  Minutiae1finger
• Proprietary-Minutiae algorithmsslightly more
  accurate than Minutiae-Minutiae
• Minutiae Record format algorithmsslightly more
  accurate than Card format

9
Main findings 2

• Different vendors minutiae templates similar
  average interoperable performance
• Different minutiae comparison algorithms
  different average interoperable performance
• Relative interoperability similar across
  databases(e.g. least interoperable pair same for
  all databases

10
Main findings 3

• Encoders
• Card format causes some problems
• Especially with Large images
• Standard does not seem to explicitly state
  maximum image size
• Matchers
• Different databases revealed issues with some of
  the algorithms
• detailed technical discussions on algorithms,
  tests, and standards to resolve these issues

11
Main findings 4

• Will need less compute intensive tests for Phase
  II
• Template generation time similar for
• Proprietary templates
• Minutiae record format templates
• Minutiae card format templates