Establishing the Quantitative Basis for Sufficiency Thresholds and Metrics for Friction Ridge Pattern Detail and the Foundation for a Standard
Randall S. Murch, A. Lynn Abbott, Edward A. Fox, Michael S. Hsiao, Bruce Budowle
See also the instructional videos on "Fingerprint Comparison"
Abstract
The purpose of this two-year project has been to address the need for a sound, quantitative basis for assessing the quality of fingerprint images. Latent prints, in particular, can be problematic because they are often partial, smudged, and otherwise distorted. Prints of sufficiently high quality routinely allow for identification (i.e., originates from one known source) or exclusion (i.e., could not have originated from a reference source). However, image quality problems related to identifiable Level 1, 2, or 3 details can be a major source of uncertainty and potential error, or may contribute to a (sometimes incorrect) determination of no conclusion. An ability to assess fingerprint image quality therefore represents a crucial step in reaching correct determinations.
The high-level goal of this cross-disciplinary collaboration has been to derive a scientific foundation for measurement of fingerprint image quality, particularly for latent prints. The objectives of this effort have been the following: to make a significant contribution to increasing accuracy, reliability, repeatability, verification, defensibility, and uniform assessment of fingerprint pattern analysis and practice; to provide a demonstrable and defensible basis for engagement of the relevant practitioner and stakeholder communities to incorporate and accept standards into friction ridge pattern analysis, reporting, and use; to provide for substantial improvements to training, proficiency testing, quality assurance, and control (quality management) that are more consistent across the forensic science community; to incorporate metrics that can be documented into the ACE-V or other accepted friction ridge examination methods; to provide the foundation for the development of novel technology aids for human examiners to automate fingerprint pattern image quality determinations; and to provide the basis for image quality determination (accept-reject) that also can be applied with automated fingerprint systems at the point of capture.
The work has been motivated in part by the Daubert ruling (Daubert v. Merrell Dow Pharmaceuticals, 1993), as well as by conclusions drawn in the subsequent study by the National Academy of Sciences, Strengthening Forensic Science in the United States: a Path Forward (2009). It is reasonable to expect scientific validity when using friction-ridge information for identification or exclusion.
The researchers on this project have followed an experimental approach, testing theoretical concepts through their application to actual images, and then performing statistical validation of the results when possible. Several image databases have been used, containing rolled prints, flat (plain) prints, and latent prints. The researchers also have obtained prints in the laboratory, using latent lifting methods as well as a dedicated live-scan imaging device. Furthermore, the researchers have digitally altered images of actual prints in order to determine drop-off points, that is, thresholds at which an area of friction ridge or feature can no longer be reliably used for identification. Metrics to quantify the effect on image quality have been developed. From these studies, quantitative thresholds have been established for unbiased selection and for use of Level 2 detail, in which both minutia and friction ridges have been incorporated into our formulation.
In conclusion, the results obtained have been noteworthy. First, our hierarchical representation of relations among minutia and friction ridges offers a unique and powerful way for fingerprint search and comparison. In addition, it allows for the mining and detection of unique and rare features that can be extremely useful when drawing statistical likelihood of a given feature. We have also been successful in developing techniques to enhance the accuracy of extraction of ridges and minutia from a print using novel filtering techniques. We developed parallel implementations of our algorithms on a low-cost general purpose graphics processing unit (GPU) and achieved a significant speed-up. Finally, we have successfully created a database of synthetic fingerprints.