Suspect identification by facial features

Abstract

Often during criminal investigations, witnesses must examine photographs of known offenders, colloquially called ‘mug shots’. As witnesses view increasing numbers of mug shots that are presented in an arbitrary order, they become more likely to identify the wrong suspect. An alternative is a subjective feature-based mug shot retrieval system in which witnesses first complete a questionnaire about the appearance of the suspect, and then examine photographs in order of decreasing resemblance to their description. In the first experiment, this approach is found to be more efficient and more accurate than searching an album. The next three experiments show that it makes little difference if the witness has seen the suspect in person or only seen a photograph. In the last two experiments, it is shown that the feature-based retrieval system is effective even when the witness has seen the suspect in realistic natural settings. The results show that the main conclusions drawn from previous studies, where witnesses searched for faces seen only in photographs, also apply when witnesses are searching for a face that they saw live in naturalistic settings. Additionally, it is shown that is it better to have two raters than one create the database, but that more than two raters yield rapidly diminishing returns for the extra cost.

Keywords:

• Suspect identification
• Face identification
• Mug shots
• Image retrieval
• Feature retrieval systems

Acknowledgements

This research was supported by grants to the first author from the National Science and Engineering Research Council of Canada and Saint Mary's University. Some of the data reported in this paper have been reported previously in various conference proceedings: IEA'94, Congress of the International Ergonomics Association, 1994, Toronto, Canada, SPIE'95, The International Society for Optical Engineering, 1995, San Diego, USA and Interact 95, IFIP Conference on Human Computer Interaction, 1995, Lillehammer, Norway.