1
Biometric Technology Today • October 2006 Biometric test results always create strong industry interest and this is especially true when the technologies under investigation are classed as ‘newer’ approaches, where typically there is no significant history of use and a lack of biometric databases from which to work. NPL and Deloitte’s recent test results studied a number of these sorts of technologies including face and iris technologies – in particular, the use of skin texture and 3D imaging in facial recognition, as well as the use of alternative algorithms and hardware developments for iris recognition. The study looked at technology and algorithms from: A4 Vision – Vision Access Identix – G6 LG – IrisAccess iCam4100 SecuriMetrics – PIER 2.3 Iris recognition algorithms from Cambridge University and SmartSensors (based on work by Don Munroe at the University of Bath). The testing took place on 210 volunteers that were provided by each of the testing organisations and the overall tests took five months – although the average time separation between enrolment and verification was three months. Offline data analysis resulted in more than 600 genuine attempts and more than 120,000 imposter attempts. The tests also measured the total time for the transaction – starting with the test subject being two metres away from the data acquisition position and ending when the system reported successful completion. Face facts Results for each individual device will be reported in full in next month’s issue of Btt. However, the overall results show that biometric offerings have developed significantly since the last trials, both in terms of hardware and software. The testers are at pains to point out that direct comparisons between different products are not meaningful unless all system requirements are known. This is because most products on the market have been developed specifically for a defined application scenario and so direct comparisons could be akin to comparing ‘apples with pears’. Face recognition was reported as experiencing strong technical advances in the last couple of years. Notably the Identix G6 engine displayed “perfect” matching performance. However, this was at the expense of greater transaction times. (Recognition took 33 seconds, compared to seven seconds for Vision Access and Iris Access iCam4100, and 16 seconds for the PIER 2.3 device.) Demonstrating just how far the systems have come, the 3D face technology supplied by A4 Vision outperformed all the biometric technologies analyzed in 2000/2001 except for the iris systems. Despite the impressive performance in skin texture and 3D imaging, the testers noted that further testing is required to assess the effects of “template ageing” over longer periods (i.e. years rather than months) for these newer approaches. Eyes front When commenting on the performance of iris recognition it was noted that the greatest improvements are related to user friendliness as opposed to performance (which was already exceptional). The results found that the LG system is quick, user-friendly and easy to use, whilst the SecuriMetrics device had with superb image quality and versatility. New iris recognition algorithms are beginning to appear on the market thanks to the expiry of the original iris encoding and matching methods developed by John Daugman at Cambridge University. The Deloitte/NPL tests investigated the performance of two new algorithms (one from Daugman at Cambridge and the other from SmartSensors) using the images captured on the two cameras mentioned above. These algorithms were compared to the default Iris2pi algorithm normally used by the two camera devices. According to the results, a multi- algorithm fusion could potentially enhance overall performance in certain cases. Quality of data As the hypothesis goes: “Garbage in, garbage out.” These tests emphatically confirmed this principle. According to the testers almost all errors recorded could be attributed to poor input data quality. NPL and Deloitte conclude that appropriate operator skills and quality control are as important as the quality of the capture device and processing algorithms. The findings prove, the testers claim, that passport photo guidelines for example are indeed necessary. In fact, they say, there might even be a merit to store images at higher resolutions to enable the future use of technologies such as skin texture. This Joint Industry Project was launched under the Measurement for Innovation programme of the UK Department of Trade and Industry (DTI). The project was funded by DTI and industry partners that included product suppliers, integrators, universities, a consultancy and the UK’s Home Office. Next month we will publish the full testing summary. NEWS FEATURE Biometric technologies impress in UK tests In the year 2000, the UK’s National Physical Laboratory (NPL) tested the performance of a variety of biometric systems. Six years on, the resulting report is still cited as indicative of biometric system performance, despite obvious advances in the technology. New tests performed by NPL and Deloitte show there has been a dramatic improvement in the performance of biometric sensors and algorithms, as well as in the development of standards for biometric data interchange and testing. This news feature outlines the key findings of the research and will be followed next month by an in-depth synopsis of the findings. 8

Biometric technologies impress in UK tests

  • View
    212

  • Download
    0

Embed Size (px)

Citation preview

Biometric Technology Today • October 2006

Biometric test results always create strong industry interest and this is especially true when the technologies under investigation are classed as ‘newer’ approaches, where typically there is no significant history of use and a lack of biometric databases from which to work.

NPL and Deloitte’s recent test results studied a number of these sorts of technologies including face and iris technologies – in particular, the use of skin texture and 3D imaging in facial recognition, as well as the use of alternative algorithms and hardware developments for iris recognition.

The study looked at technology and algorithms from:• A4 Vision – Vision Access• Identix – G6• LG – IrisAccess iCam4100• SecuriMetrics – PIER 2.3• Iris recognition algorithms from

Cambridge University and SmartSensors (based on work by Don Munroe at the University of Bath).

The testing took place on 210 volunteers that were provided by each of the testing organisations and the overall tests took five months – although the average time separation between enrolment and verification was three months. Offline data analysis resulted in more than 600 genuine attempts and more than 120,000 imposter attempts. The tests also measured the total

time for the transaction – starting with the test subject being two metres away from the data acquisition position and ending when the system reported successful completion.

Face factsResults for each individual device will be reported in full in next month’s issue of Btt. However, the overall results show that biometric offerings have developed significantly since the last trials, both in terms of hardware and software.

The testers are at pains to point out that direct comparisons between different products are not meaningful unless all system requirements are known. This is because most products on the market have been developed specifically for a defined application scenario and so direct comparisons could be akin to comparing ‘apples with pears’.

Face recognition was reported as experiencing strong technical advances in the last couple of years. Notably the Identix G6 engine displayed “perfect” matching performance. However, this was at the expense of greater transaction times. (Recognition took 33 seconds, compared to seven seconds for Vision Access and Iris Access iCam4100, and 16 seconds for the PIER 2.3 device.)

Demonstrating just how far the systems have come, the 3D face technology supplied by A4 Vision outperformed all the biometric

technologies analyzed in 2000/2001 except for the iris systems.

Despite the impressive performance in skin texture and 3D imaging, the testers noted that further testing is required to assess the effects of “template ageing” over longer periods (i.e. years rather than months) for these newer approaches.

Eyes frontWhen commenting on the performance of iris recognition it was noted that the greatest improvements are related to user friendliness as opposed to performance (which was already exceptional).

The results found that the LG system is quick, user-friendly and easy to use, whilst the SecuriMetrics device had with superb image quality and versatility.

New iris recognition algorithms are beginning to appear on the market thanks to the expiry of the original iris encoding and matching methods developed by John Daugman at Cambridge University.

The Deloitte/NPL tests investigated the performance of two new algorithms (one from Daugman at Cambridge and the other from SmartSensors) using the images captured on the two cameras mentioned above. These algorithms were compared to the default Iris2pi algorithm normally used by the two camera devices.

According to the results, a multi-algorithm fusion could potentially enhance overall performance in certain cases.

Quality of dataAs the hypothesis goes: “Garbage in, garbage out.” These tests emphatically confirmed this principle. According to the testers almost all errors recorded could be attributed to poor input data quality. NPL and Deloitte conclude that appropriate operator skills and quality control are as important as the quality of the capture device and processing algorithms.

The findings prove, the testers claim, that passport photo guidelines for example are indeed necessary. In fact, they say, there might even be a merit to store images at higher resolutions to enable the future use of technologies such as skin texture.

This Joint Industry Project was launched under the Measurement for Innovation programme of the UK Department of Trade and Industry (DTI). The project was funded by DTI and industry partners that included product suppliers, integrators, universities, a consultancy and the UK’s Home Office. Next month we will publish the full testing summary.

NEWS FEATURE

Biometric technologies impress in UK testsIn the year 2000, the UK’s National Physical Laboratory (NPL) tested the performance of a variety of biometric systems. Six years on, the resulting report is still cited as indicative of biometric system performance, despite obvious advances in the technology. New tests performed by NPL and Deloitte show there has been a dramatic improvement in the performance of biometric sensors and algorithms, as well as in the development of standards for biometric data interchange and testing. This news feature outlines the key findings of the research and will be followed next month by an in-depth synopsis of the findings.

8