Date of Graduation
Statler College of Engineering and Mineral Resources
Lane Department of Computer Science and Electrical Engineering
Keith B. Morris.
The definition of standards in fingerprint identification is currently an issue of discussion in the field. Quantitative standards have been used in the past to provide justifications for conclusions; however, a scientific basis for relying on numerical standards alone currently does not exist. The tradeoff for this combined approach is that conclusions are based on a conclusion that is left to the judgment of the examiner and may not be repeatable. To test the implementation of thresholds for conclusion, this research studied the effects of only considering concrete data in quantitative form. In this case, signal detection theory is applied to latent fingerprint matching by using automated fingerprint identification systems from two different program vendors. By searching a test set of fingerprints multiple times with a wide range of detail entered, values for the number of system-matched minutiae and computed match scores can be studied to determine threshold limits based on the amount of the search returns. This in turn allows for the generation of receiver operating characteristic curves that directly measure the reliability of the system. The results show that the ability of the system to distinguish matches and non-matches properly is partly based on the method by which the searches are evaluated. Furthermore, the searched area of the fingerprint and the size of the database play roles in determining how well the system is able to discriminate between states. Through future comparison against results submitted by latent fingerprint examiners, inferences can be drawn as to the reliability of conclusions based on varying levels of available detail.
Widman, Eric J., "Quantifying the Reliability of Latent Fingerprint Matching via Signal Detection Theory" (2011). Graduate Theses, Dissertations, and Problem Reports. 3329.