Recognizability of Demographically Altered Computerized Facial Approximations in an Automated Facial Recognition...

Area of Science:

• Forensic Science
• Computer Vision
• Biometrics

Background:

• Facial approximations are crucial for identifying unidentified persons.
• Demographic alterations can impact facial approximation recognizability.
• Automated systems require robust facial recognition capabilities.

Purpose of the Study:

• To evaluate the recognizability of demographically altered facial approximations.
• To assess the utility of these approximations in unidentified persons tracking systems.
• To compare the impact of sex versus ancestry alterations on identification rates.

Main Methods:

• Generated five computer-based facial approximations for 26 African male participants.
• Altered demographics included: African male (true), African female, Caucasian male, Asian male, and Hispanic male.
• Performed automated blind searches against a standardized gallery of 6159 photographs.

Main Results:

• True demographic approximations achieved a 62% identification rate.
• Sex-altered (African female) approximations yielded a 50% identification rate.
• Ancestry-altered approximations showed lower identification rates: Caucasian (42%), Asian (35%), Hispanic (27%).

Conclusions:

• Facial approximations altered to the opposite sex are operationally informative when sex is unknown.
• Ancestry-altered approximations are less effective than sex-altered ones for identification.
• Sex alterations in facial approximations offer more constructive data for tracking systems than ancestry alterations.