Meeting the Challenges of Garment Comparison

Criminal surveillance vs. garment comparison image

An Increasing Problem in Forensic Analysis

Forensic examiners increasingly analyze audio and visual evidence associated with terrorist and criminal investigations. Because video cameras are prolific, investigators can compare physical items recovered from suspects with those depicted in surveillance system footage. However, several factors hinder their efforts:

Therefore, it is unclear how to assert quality of match (e.g., strength of evidence) in a consistent and principled manner, leaving courtroom testimony susceptible to challenge.

There Is a Solution

Measurements and probability of false match calculations image

The Department of Homeland Security Science and Technology Directorate sponsored Quantum Signal, LLC (QS) to develop a method to statistically analyze the forensic comparison of patterned garments. In prior work, QS researchers showed that statistical upper bounding of the incidence rate (i.e., the worst case probability of accidental match) may be performed in cases involving U.S. military camouflage uniforms by measuring the relative alignment of the camouflage pattern on different garment pieces. In cases where the forensic examiner concludes that the seized garment would be indistinguishable from the surveillance image if worn by the perpetrator, the likelihood ratio (LR) is approximated by the reciprocal of the incidence rate.

Capitalizing on this work and its unique expertise in statistical analysis and in signal and image processing, QS performed an empirical study to develop explicit guidelines for determining if statistical evaluation of pattern shifts is an appropriate technique for a given garment comparison case and, if so, how to achieve a statistical upper bound on the incidence rate (and thus a lower bound on the LR). QS published the results in a technical report and created the Garment Match LR Calculator.

Bar Graph With Normal Curve