Eyewitness ID: Part II The Composite Sketch

Image courtesy of The Innocence Project

In Eyewitness ID: Part I, we recapped the story of Ronald Cotton who was wrongfully convicted based on a bad eyewitness identification. Jennifer Thompson assisted police in drawing a composite sketch which was later used in creating the line-up in which Mr. Cotton was mistakenly chosen. Ms. Thompson stated,

When I went to the police department later that day, I worked on a composite sketch to the very best of my ability. I looked through hundreds of noses and eyes and eyebrows and hairlines and nostrils and lips. Several days later, looking at a series of police photos, I identified my attacker. I knew this was the man. I was completely confident. I was sure.

The composite sketch was one link in the chain that lead to Ronald Cotton's wrongful conviction.

The problem with the composite sketch is that, like the name implies, the perpetrator is constructed from a composite of features. The eyewitness either describes the perpetrator to a skilled artist or reconstructs the face jigsaw-like from a selection of facial features. This cuts against human psychology which supports the notion that it is easier to recognize a whole face.

The solution? As an article in Forensic magazine explains: "Two research groups in the UK independently stumbled upon similar solutions as to how to produce a better “facial composite” of a suspect in a more psychologically natural manner. Peter Hancock and Charlie Frowd at the University of Stirling, Scotland, have produced a system called EvoFIT while Chris Solomon at the University of Kent, England, has developed a similar system called EigenFIT." The science, briefly, involves Genetic Algorithms-the approach begins with a set of random solutions. The best solutions are selected and then “bred” together. As explained by EvoFIT:

EvoFIT begins by creating a set of faces (example screen) with random facial shapes and facial textures. A witness would normally select six of these shapes and textures that most resemble a suspect. These selections then become the “parents” of the next population; to produce another generation, the components of the selected faces are mixed together. The “offspring” faces are selected and bred together as before. The selections enable the set of faces to become more like the suspect. Evolution is completed when an acceptable likeness is found; the relevant image is then saved to disk as the composite.

Given that the software depends on the accuracy of the first faces chosen, Dr. Frowd admits that the genetic algorithm is also capable of producing serious errors:

One drawback of genetic algorithms generally, is that they can converge on the wrong solution if the initial population is poorly chosen. “This may happen,” Dr. Frowd admits, “but if it does, the system is rolled back and started again.”

That takes us back to where we began, the fallibility of a witness's memory...


Example of EigenFIT courtesy of eyeID.