home

Sketch Released of Man Sought in Judge's Family Killings

Someone should drop a dime on this guy.

If he's not involved, the police need to know so they will keep looking for the murderer(s) of Judge Joan Humphrey Lefkow's husband and mother. If he is involved, he needs to be charged and tried.

I'm not a big fan of composite sketches, particularly those created by software programs and Identi-kits. Eyewitness guru and Professor Gary Wells has a new manuscript which discusses multiple past research studies that have found that participating in the creation of a sketch can impair the witnesses' memory of the perpetrator. Wells recently conducted two new studies and reaches the same conclusion.

The programs currently in use by law enforcement utilize a "particularized" approach...the witness picks out individual facial features which are then combined into a face. This is counterintuitive to the memory process, which operates on a more holistic level.

Another reason that composite systems may result in poor matches to original faces is the principle of "blended memories."

... it is possible that building a composite face creates a second memory; there are now two memories of the face, the original memory and the composite memory, and they compete in any later memory task. ....it is [also] possible that the original memory is blended or averaged with the original memory, yielding a new face that has some characteristics of the original face and some of the composite face.

More possibilities:

It is possible that the original memory of the face is replaced with the composite face, yielding only one face in memory, namely that of the composite. There are also hybrid possibilities that involve more than one of these processes.

I think also that the more the sketch is publicized, the more likely the witness will be to compare his or her memory of the perpetrator to the sketch rather than to the person they saw during the event.

Nonetheless, Wells says that composite sketches may retain some value in criminal investigations. He concludes:

On the one hand, we know that composites usually produce results that poorly resemble the actual face that they were meant to depict (citations omitted) and that composites can apparently damage the memory that the witness has of the original face (current research). On the other hand, we do not know how often composites somehow manage to help crime investigators eliminate potential suspects or narrow the search of possible suspects. Furthermore, in multiple witness crimes it might be possible to use one witness to build a composite and save the other witnesses for any later lineup identification attempts.

Wells cautions that his findings may not be applicable to police sketch artists.

Furthermore, we cannot be certain that the effects we observed here are applicable to sketch artists. Like computer and transparency-based face composite systems, the sketch artist process is somewhat featural based rather than more purely holistic. Unlike computer and transparency-based face composite systems, however, the sketch-artist process does not require the witness to examine isolated facial features and select among them. Hence, we reserve judgment on the sketch artist process until we more fully understand the processes leading leading to the effects that we observed here.

So yes, be on the lookout for the man in the composite sketch. Report him to police if you see him. But don't presume he is the murderer. Whoever committed the atrocious killings of Mr. Lefkow and Mrs. Humphry needs to be apprehended, sooner rather than later, but if police nab the wrong guy, the killer(s) remain at large, able to kill again.

< Mayor Oscar Goodman Tells the Truth | Police Probe Judge's Relatives Murders >
  • The Online Magazine with Liberal coverage of crime-related political and injustice news

  • Contribute To TalkLeft


  • Display: Sort:
    Re: Sketch Released of Man Sought in Judge's Famil (none / 0) (#1)
    by Ray Radlein on Wed Mar 02, 2005 at 10:25:54 PM EST
    It occurs to me that one might be able to arrive at a composite sketch through an entirely holistic approach, by using a genetic algorithm: That is to say, have the computer generate two entirely random faces (perhaps within certain very broad parameters, just to save some time) and ask the witness which face looks more like the perpetrator. Then do it again with two more faces, and again, and again — eventually, through Bayesian inference, the computer will begin to converge towards the face of the perpetrator; and since the witness is being presented with an entire face to consider at each step of the process, their internal comparisons will be done in exactly the same way that the brain usually recognizes faces. I would be willing to bet a large amount of someone else's money that a system like that would yield better results than a standard identikit; and it's certainly well within the capabilities of modern technology (heck, the software could even make the faces almost photorealistic at each stage, although it's possible that cartoony faces might actually be better than realistic ones for this purpose).

    Re: Sketch Released of Man Sought in Judge's Famil (none / 0) (#2)
    by Ray Radlein on Wed Mar 02, 2005 at 10:27:21 PM EST
    In fact, for what it's worth, I am so certain that this would work that I would be flabbergasted if I learned that it isn't already being tried somewhere. I can't be the only person to think of this.

    Re: Sketch Released of Man Sought in Judge's Famil (none / 0) (#3)
    by Lora on Thu Mar 03, 2005 at 06:02:15 AM EST
    Here's another thought...how about having the witness attempt his/her own drawing first, regardless of what kind of artist the witness is...so as not to 'contaminate' the memory with other faces? I'm a lousy artist, but I think I'd want to try that first, if it was me. Some features might be outstanding in my memory, and I'd like to get them down on paper first, myself, then get help from an artist to improve/refine the picture. I would also probably make a written description first, perhaps, as I'm better with words than drawings, as precisely as I could.

    Re: Sketch Released of Man Sought in Judge's Famil (none / 0) (#4)
    by Sailor on Thu Mar 03, 2005 at 07:20:51 AM EST
    Ray - I'll take that bet. Psychophysics has shown that that approach is worse because each iteration 'colors' the original memory and blends with the previous iteration. Facial recognition is a very special part of brain function and fMRI scans show most of the procesing takes place in a single area of the brain. We 'match' faces to previously known ones. It isn't a 'puzzle fitting' operation but a 'flash card' type. The best idea I've seen so far is to have about a dozen actual faces that meet the original description with certain parameters (e.g. eye width, chin size, nose flair ... etc.) varied. The all 12 are shown at once and the witness picks the first one that comes to mind.

    Holistic methods may lead to better pictures -- a strong reason to prefer them -- but I seriously doubt that any method will prevent considerable contamination of the witnesses memory with the images he is exposed to during the process of generating the "sketch".

    Re: Sketch Released of Man Sought in Judge's Famil (none / 0) (#6)
    by Sailor on Thu Mar 03, 2005 at 09:37:00 AM EST
    cmdicely - you are correct, but I was specifically talking about a method for generating an image of a suspect where one didn't previously exist. This is a different case than corruption of a eye witness memory for court purposes. The prosecution has to make a decision on whether 'tainting' a witness like this is worth it. Much in the same way they decide whether to use hypnosis. Please understand, I am not speaking to the legal ramifications, just the scientific ones. We don't process vision like a TV screen where the whole picture is constantly renewed. Almost all of adult vision is pattern matching and motion sensing and our brain puts pictures up that are previously stored. That's how a lot of optical illusions work; there will be 2 solutions to a given image and we can flip back and forth between them.

    You know that maybe Bush? not joking.