My research project for the dissertation component involved the analysis of three facial composite systems: EVOfit, Facefit and PRO-fit. A facial composite is a likeness created of a person of interest in a crime by either a witness or a victim under the guidance of a Forensic Artist. Previously, all facial composite systems were feature-based and involved the selection of individual facial features, sometimes in isolation to a face. In the last decade new types of systems have emerged that focus upon "holistic" creation and focus on viewing whole faces such as EvoFIT. This method is thought to align better with how our memory processes unfamiliar (stranger) faces. Using an algorithm, the system learns from the witness/victims' selections and generates an array of new faces based on the parameters of the last selection as well as a unique set of Holistic Tools that alter the final face. This allows the witness/victim to slowly narrow down the likeness until they believe it closely matches the person of interest.
To test the accuracy of classical and modern facial composite systems I conducted interviews with "witnesses" who had viewed a target face for 30 seconds before returning 1-2 days later to produce a facial composite. There were ten target individuals who were selected from the British television show EastEnders and a total of 30 composites were produced of them (one for each system). After completion of the composites, members of the public and of my affiliated universities (University of Dundee and University of Central Lancashire) were invited to participate in an online questionnaire to identify the composites. This questionnaire involved two stages: the first invited the participants to name the composites and the second showed the real image of the target individual for naming. The second stage was crucial to the analysis as an individual who cannot name the target individual would be unable to name the facial composite and thus was excluded from the statistics.
My analysis concluded that there was no significant difference in the performance of EvoFIT or Facefit with both systems scoring approximately 22% correct identifications. However, it did show that the classical system, PRO-fit, performed significantly worse with only ~8% of facial composites identified. As a feature-based system, Facefit has outperformed all other tested feature-based systems before it and more research is needed to discover why
Finally, all of my work was presented in an exhibition which included interactive elements that explored the history and difficulties of facial identification alongside a selection of 10 facial composites from the project which were set up to allow viewers to try their hand at naming.