CS909 - St: Biometric Identification

Intructor: Thomas Cahill
Fall 2004 Two Sections: Mon. 6:00-6:15PM(I) & Wed. 6:00-8:15PM(I2)

This Computer Science course will be an interdisciplinary learning experience Seniors and Graduate Students in the areas of Computer Science, Computer Engineering, Electrical Engineering and Management. It will use the concept of "Biometric Identification" as an exploratory paradigm for the analysis of the interaction of state of the art technology with the process of invention as well as the real world processes of technological innovation.

Course Features:

1. The course is done "live" on the wireless laptop for direct projection within the classroom. Extensive use of "Matlab" will be made to display mathematical processes and demonstrate actual signal processing.
2. The class will have its own yahoo eGroup called "biompolyu02" which each student will be required to join. The group has mailing list services, file upload/download capability, group polling and chat.
3. Tracking: Special Tracks will be established for a) Computer Scientist/Management, b) Computer Engineering and c) Electrical Engineering/Graduate AND for i) Voiceprinting, ii) Face Recognition iii) Fingerprint and iv) Iris Scan.

All students will be responsible for doing some assignment(s) in MATLAB. Students in each track will receive at least one unique question on the Midterm to be answered by them alone.

4. Challenge Projects: Especially well prepared and confident students will have the opportunity to work on a "challenge project." Successful completion of a challenge project should produce results that will be publishable in standard technical media and/or reducible to patents.
   Voice Identification Challenge Project: In 1979 the National Academy of Sciences found that there was no evidence to support that Interspeaker Variability was greater than Intraspeaker Variability. If this is not so, the whole "voice identification agenda" collapses. Challenge - have the intervening decades produced positive evidence in either direction relative to the NAS findings? Which current or proposed voice-id technologies claim best performance? Can we determine that voice-id is in general impossible?
   Iris Challenge Project: Current Iris Identification systems are based on a US Patent issued to John Daugman. Two years ago students in Texas attempted to create an implementation of Daugman's system (using Matlab). They found that it was impossible to implement the system described by Daugman because certain critical parameters were missing. Challenge - build Daugman's system in Matlab, C++ or a mixture of both. If you confirm that the Daugman patent specifications are incomplete - that would argue for the invalidation of the patent. If you find the "secret parameters" that make it work or improve on test performance, you should be able to patent that and have a major stake in the Iris ID field.
   Fingerprint Challenge Project: Fall 2000 a contest was held among Fingerprint Verification Algorithms called FVC2000. A small Lithuanian company, Neurotechnologija Ltd., won "hands down" with an equal error rate (EER) 1.37%. Challenge - validate the Neurotechnologija results from an independently produced database of fingerprints. Determine why they did so well. Improve on their method, if possible. Analyze those results in view of the FVC2002 results which will be released during the course.