AMES, Iowa -- In the latest high profile investigation, an eyewitness composite sketch of the woman suspected of stealing a 10-day-old infant and slashing the mother's throat in Missouri was used. But prior research has found that facial composites of criminal suspects built by individual eyewitness accounts often produce poor likenesses of the actual perpetrators.
Now a new study by Iowa State University Distinguished Professor of Psychology Gary Wells and one of his Ph.D. students reports that the morphing together of composites made by four individuals produces an image that better resembles a target face than an individual composite, as long as the morphed composites don't become too numerous.
"The problem with composites is that frankly, they don't look much like the person," said Wells. "It's a difficult task for someone to come up with a composite because we don't perceive faces piecemeal -- we don't store them in memory as individual facial features, such as the nose, the eyes, or the mouth. Rather, the facial processing is holistic. So in asking someone to come up with a composite, we're asking people to do something that is totally unnatural."
Lisa Hasel, an ISU doctoral student in psychology, joined with Wells to author a research paper titled "Catching the Bad Guy: Morphing Composite Faces Helps," which has been accepted for publication in Law and Human Behavior, a professional journal. The study was awarded a national prize by the Association for Psychological Science.
The methodology
The researchers selected 16 target faces from a large database of facial photographs. Each was of 19- to 23-year-olds that fell into four different categories: four Asian females with long black hair, four Caucasian females with short brown hair, four Caucasian males with short brown hair, and four Caucasian males with medium-length black hair.
Sixteen Iowa State students participated in the composite generation phase of the study, with 72 more participating in an initial attractiveness evaluation phase, and another 60 in the similarity evaluation and second attractiveness evaluation phase.
In the composite generation phase, participants were asked to look at one of four different target faces in a specific category for one minute. After each face was removed, they were asked to create four composites apiece using the facial composite software called FACES. After building each composite, the creator rated their own compositions for how similar they thought their composite was to the face they had viewed. The four composites for each face were then morphed together by an independent experimenter -- resulting in 64 individual composites and 16 morphed composites.
Participants in the next phase rated the attractiveness of every face image, comparing the overall attractiveness of the three face image categories -- photos of target faces, composites, and morphs. The morphed composites were found to be the most attractive of the three.
The morph effect
The final group also rated the attractiveness of the face images, as well as the similarity of the individual composites and morphs to both the target and non-target faces. Upon completing their ratings, they were asked to select from a face array, which included the target face, which face looked the most similar to the individual composites or morphs. The morphs produced a higher target-preference rate (47.7%) than did individual composites (35.2%).
"Does morphing four composites into one represent a specific person better than individual composites? The bottom line is that the answer is 'yes,'" said Wells.
But Wells also reported the following two side-effects associated with that conclusion:
1. The morphed image was found to be more "attractive" than the individual target faces.
"So if the perpetrator is an unattractive person, the morphed image may not work as well because the resulting face may be too attractive," said Wells.
2. The morphed image better resembles the intended target face, but also better resembles other faces who are not the intended target.
"If you morph a number of these composites together, you ultimately create a closer image to the prototypical face," Wells said. "So if you put 10 composites together from different people of the same types of face -- for example, white males -- they would produce a morphed image that would look like the average white male."
"We just went up to four in this study, and it does produce a better likeness of the intended face, although it also does start resembling other faces in that general category," he said. "It ultimately would produce more leads, but also a greater chance of including the perpetrator of the crime."
According to Wells, this is the first real improvement to the tool of eyewitness composites in a criminal investigation.
A copy of the paper is available through Wells' Web site at: http://www.psychology.iastate.edu/FACULTY/gwells/homepage.htm.
