No open-access location is known for this record yet.
No open-access location is known for this record yet.
Purpose. Previous laoratory‐based research suggests that facial composites, or pictures of suspected criminals, from UK computerized systems are named correctly about 20% of the time. The current work compares composites from several such systems following a more realistic interval between seeing an ‘assailant’ and constructing a composite. Included are those used by police in the UK (E‐FIT, PROfit and sketch), and the USA (FACES), and a system in development (EvoFIT). Method. Participant‐witnesses inspected a photograph of a celebrity for 1 minute and then 2 days later constructed a composite from one of these systems using a procedure closely matching that found in police work; for example, the use of a Cognitive Interview and computer operators/artists who were appropriately trained and experienced. Evaluation was assessed mainly by asking independent observers to name the composites. Two common auxiliary measures were used, requiring composites to be matched to their targets (sorting), and photographs to be chosen from an array of alternatives (line‐up). Results. Composite naming was surprisingly low (3% overall), with sketches named best at 8%. Whereas composite sorting revealed a broadly similar pattern to naming, photo line‐ups gave a poor match. Conclusion. With a 2 days delay to construction, the results suggest that, while likenesses can be achieved, few composites would be named in police work. The composite sorting data provide further evidence that the computerized systems tested perform equivalently but are poorer than the manually‐generated sketches. Lastly, the data suggest that line‐ups may be a poor instrument for evaluating facial composites.
Loading…
Stored only when both papers are in the corpus — 103 citations exist upstream.
Papers that cite the same references — the relational fallback while embeddings are still backfilling.
Imported from OpenAlex (CC0) — nothing here is generated.