Remember this face?

Computerized caricatures might jog witnesses’ memories better than sketches

LAS VEGAS SUN

This collection of computer composites, top row, compared with photos of the same people — bottom from left, Britain’s Prince William, David Beckham, Ben Affleck and President Bush — shows the results of morphing dozens of computer-generated faces to reach a likeness of an individual face a witness can identify.

By Abigail Goldman (contact)

Saturday, March 1, 2008 | 2 a.m.

Boulder City Police hope sketch provides answers in robbery, rape (5-20-1997)
Investigators interview witnesses, try to compile sketch of killer (3-12-1997)

Caricatures are more effective than police sketches (11-26-2007)
Doing Police Sketches: 'Not an Exact Science' (6-11-1996)

Composite sketches, those shaded pencil drawings of shadowy crime suspects plastered in post offices and police stations, are, as it turns out, practically useless.

At least according to a group of British academics who say composite sketches lead to successful identifications only 10 percent of the time, and that is in lab tests in which the drawn subject is actually known to the person trying to make the identification.

When suspect sketches are released in the real world and it’s up to Joe Citizen to put a name to the face, the identification rate drops into single digits, says Charlie Frowd, a professor at the University of Central Lancashire in England.

It’s a disappointing statistic that scientists say reveals a truth about how we see and remember faces — not as a collection of separate features that a sketch artist might ask witnesses to recall, but as a whole greater than the sum of its parts. We don’t remember a nose and then eyebrows and then a chin; we remember them all at once, academics say, and in relation to one another.

This means the way police artists create suspect composites — whether drawn or compiled from photographs of facial features, as some police departments do — is fundamentally flawed.

Just look at the sketch of the Unabomber that came out before Theodore Kaczynski was caught — the sunglass-wearing, mustachioed, curly-haired suspect was so far from the grizzled Kaczynski that it was laughable, said Gary Wells, an Iowa State University psychology professor who studies the reliability of eyewitness identification.

But the flaws in the science of sketches and suspect identification are serious, Wells said. Of the dozens of DNA prison exonerations this country has allowed, about 75 percent of the convictions were cases of mistaken identification, he said.

The solution, Frowd and fellow academics say, is caricatures. Not those hackneyed carnival drawings sold to canoodling couples eager to see themselves grotesquely exaggerated in ink, but a precise scientific caricature, a compilation of dozens of faces, slowly morphed into one that best fits the witness’s memory.

It seems logical that we naturally remember the unusual features of a face more clearly — a pair of big ears, wide nostrils, furry eyebrows. Showing a composite with these features exaggerated seems to jog the memory better than a typical composite, Frowd said. But we each remember these unusual features to a different degree. Frowd’s most effective composites are actually animated — they move through a series of caricature exaggerations and sort of undulate back and forth on a computer screen, so the face seems to change from one person into another. The nose moves up and down, the face widens and then tightens, the forehead rises and falls, the eyes expand and relax.

Frowd found that by showing study subjects a face being progressively caricatured in this manner, not in a set of exaggerated steps but as a gradual movement, he was able to bump the recognition rate by 20 percent on average, and sometimes higher.

Study subjects reported that at one point in the progressive caricature, the identity of the face seemed to “pop” out at them, and they suddenly knew who it was, Frowd said. In other words, the caricatured face had morphed to a degree that triggered the subject’s memory of the person being identified.

Now Frowd is promoting this undulating composite under the name EvoFIT, and a police department in Lancashire has used the software to catch a wanted burglar, he said.

The sketch artist who works for Metro Police would not comment to the Sun, unwilling to weigh in on the controversy surrounding composite techniques.

Frowd’s system works by presenting a witness with 72 slightly different male or female faces, depending on the suspect. The witness picks six faces that most closely resemble his remembered face. The EvoFIT program then “breeds” those six faces to make another 72 faces, and the witness picks six from this new set. The process repeats for a third and final time, and the witness is left with six blended faces, one of which he picks as closest to what he remembers.

The final chosen face can be progressively characterized, so that the image can resonate for more than the single witness.

More often than not, police departments come up with a composite image to appease the media and the public, which often clamors for an image of a suspect, Wells said.

For some witnesses, however, the mere act of working with a composite artist can skew their actual memory of the suspect, Wells said. When witnesses are asked to work through a face feature by feature, the inherently unnatural process of feature recall can slowly replace their actual memory.

“When you are done, there is now some kind of new impression you have of that face,” he said. “The composite becomes a new memory that competes with the old memory or somehow replaces it or alters it.”

This actually gives a degree of cover to the real perpetrator, he said. It’s one of the reasons Wells wants law enforcement to use composites only as a last resort.

Metro Police’s most recently released composite sketch is of a man allegedly named “Bob,” wanted in connection with a hit-and-run, alleged to have been intentionally fatal, that police say might have resulted from a fight at the Dew Drop Inn in December. Police released two pencil sketches of Bob, one of which showed him wearing glasses and having a facial scar.

The case is still being investigated.

Discussion: 3 comments so far…

By FAinCA

3/1/08 at 7:39 a.m.

Suggest removal

I was disappointed to read this article demeaning hand drawn composites. Would it be possible to know how this article-advertisement was generated and by whom? It is my understanding that the caricature and morphing studies have not been completed using everyday individuals and is still at an experimental stage with regard to Evofit in use with police and Goldman did only mention one practical success. I was surprised to see the skewed references in the extremely dated sidebar archives leaning toward this method as though trying to verify the story. A sketch artist always works toward the overall "look" of the subject which is a similar impression these caricature and morphing programs are trying to project. Hand drawn comps have been used to much success since the 1800's and the one sure way of not having them identified is to NOT show them. At least this lop-sided article could have been redeemed by showing "Bob's" face in the media one more time with hopes of taking him off the street.
By stahld

3/1/08 at 1 p.m.

Suggest removal

I was glad to see that all of the misguided assertions in the first few paragraphs of the article at least had a purpose, which apparently is the sales of a product. There are a number of flaws readily apparent to the experienced forensic artist here. The most glaring one is that the system or its operator select faces and ask the witness to choose from among them rather than asking the witness to describe something that is already in their memory. I would think that it might force a witness into settling for something that is close, rather than constructing an image that is accurate. The "academics" who conducted this study may have found it relatively easy to have their participants achieve a likeness of a celebrity whose face is frequently seen in the media, but perhaps they should have made the attempt to achieve the same results with a subject who is seen in a fleeting moment under a situation of stress. My years in law enforcement as an investigator have demonstrated that the two situations are entirely different.

I acknowledge that software such as this has some potential value if used by a person trained in the interview of witnesses or victims of crime, but to present it in this manner is irresponsible. I find the one-sidedness of the article disturbing, and I find the likeness they achieved, even of these well known faces, to be of little value. Perhaps you should ask readers to guess at the identities of the composites or, as you call them, caricatures, without revealing the photos of the celebs.
By thecopartist

3/4/08 at 10:07 a.m.

Suggest removal

I am now retired but was one of those "shaded pencil artist of shadowy figures" for over 20 years. I was not against trying something new, if it would legitimately help in identifying a suspect but haven't found anything better than having an artist trained in interview techniques and able to use the information gathered to do a composite drawing. Granted, some composites that have been used across the country are a little strange but many departments will use an officer that can "draw" to do a composite, with little or no formal training.

Those that have had good,formal training in composite (forensic) art have a remarkable success rate in identifications.

The article, to me, seemed to be more advertisement for this guy's program (that isn't free, I'm sure) than giving the facts of forensic art. There have been many computer programs that have come and gone over the years and departments either stay with or go back to the pencil pusher.

What people have to understand is whether you use a computer or a pencil, the artist is limited to the recall abilities of the witness and some witnesses just aren't that good but you work with what you have. Believe me, I have done many a drawing from a witness that describes a suspect as a tall, short, skinny, fat man with no hair that turns out to be a female with black hair!

But it's the artist that gets blamed for a bad composite and not the witness we had to deal with. And no computer program is going to overcome that.

I'll stick to my pencil, thank you.

« Previous Ten Next Ten »

Live chat: Tuesday, noon PST; Chat with Krista Creelman; Problem Gambling Center executive director Krista Creelman will answer questions about gambling addiction from Las Vegas Sun readers from noon to 1 p.m. Tuesday, Nov. ... Submit question