I thought I had written on the problems with the Smithman e-fits before, but I cannot find the relevant post. My apologies if this duplicates something earlier.
I wanted to see if the software used to construct the two Smithman e-fits could be identified, given that the quality of the two photos is quite dissimilar. I was hoping from that to get a bit more insight in the methods used by the private investigators to build these. As it turns out, I needn’t have bothered, as the real story lies elsewhere.
Here are the two Smithman e-fits.
Note one is high quality while the other is rough. It is unlikely that both were constructed by the same operator using the same software, though not impossible, as one potential step is enhancing the e-fit using some sort of graphics package.
The e-fits were produced sometime during the employment of Oakley International, according to Henri Exton, who worked on the case at that time. This employment started in March 2008 and ended in September 2008, with the final report, including the e-fits being handed over in November 2008.
So what software was available at the time? In the UK, the packages used by the police were E-FIT and PRO-fit. In the US, they were FACES, Identikit 2000, ComPhotofit, CRIMES, Compusketch, CD-FIT, E-FIT, and Facekit. That is quite a lot of choice, The source that identified these mentioned that it had an example of the results of each of these, but unfortunately it did not. And on my travels I found the names of other packages in use at the time.
The academics were keen to know how effective these tools were and the best methods of deploying them, and it turns out it is far from easy. The following was all known in 2007.
First the operator has to be conversant with the software itself and alternatives it provides.
Second, the operator should know that trying to get a verbal description of facial features is a poor way to go about it. Apparently we process faces as a whole, not as individual parts. So a better method is to start off with a general face, then gradually evolve it until it looks sufficiently accurate.
Third, the operator should know that we tend to recognise faces more from the top half than the bottom half. Oddly, it seems hair style is the most important feature.
Fourth, the operator should know we process faces differently if they are familiar or unfamiliar. This could well be an issue after the blanket media coverage of Gerry McCann’s image.
Fifth, the operator should understand the interaction techniques known as Cognitive Interviews, Context Reinstatement and Conversational Management. Let’s skip these.
I am assuming that as the people in Oakley International were UK-based, one of the two systems used by UK police was deployed. The higher quality e-fit looks like the output from E-FIT, while the lower quality one looks like it came from an earlier version of the same software. Of course, Oakley International could have sub-contracted this work to someone who was familiar with one of the other packages.
Whether Oakley International or a sub-contractor had the relevant capability to produce the e-fits claimed by Henri Exton is a matter of conjecture, but it turns out not to matter.
In December 2007, an article appeared in the journal Ergonomics. The authors said the efficacy of e-fits had been widely studied in the UK, and they ran a study to compare package E-FIT (used in both the UK and the US) against other packages used in the US. The relevant finding is the concluding sentence of their abstract. “The results support previous findings that modern systems do not produce identifiable composites.”
This, in my opinion, is a fairly damning indictment of 2007 capability. Please note, UK universities have continually developed much better software since then, but that is of limited relevance to the Madeleine McCann case.
The other thing that came across re 2007/2008 studies was the methods used. To give but one example, a study was done as follows. Volunteers were given a photo of someone they had not seen before, an unfamiliar face. The photo was high quality and well lit. They were allowed to study it for one minute (60 seconds). Four hours later, they participated in attempting to construct an e-fit of the face in the photo. Typical success rates were 15 to 20%.
Contrast this process with the reality of the Smith sighting. The Smiths did not get 60 seconds to study anyone’s face. The lighting was not of a quality to enable e-fit construction. They did not compile the e-fit 4 hours later but some 11 to 17 months later.
The Smiths are not to blame for taking part in an e-fit construction. They were hardly to know that all the science said it was a total waste of time, and the results would be useful only if lady luck shone on the process.
Henri Exton and Oakley International should have known a bit more about a tool they were using. Perhaps they were relying on the expertise of those who produced the software, the training, and the knowledge that e-fits were in common use in the UK and US.
The situation with Operation Grange and DCI Andy Redwood is a little more puzzling. Surely by October 2013 they would have known about some of the issues surrounding those two e-fits? One assumes they had the full Oakley International file, can date the construction of the e-fits, know what software was used to produce them and have some idea of the process used.
I believe DCI Andy Redwood visited Martin Smith a couple of time before Crimewatch 2013, and I wonder if e-fit construction was part of the agenda.
I have just chunked my way through a 26 page thread on the UK Justice Forum, where a key question was when did Operation Grange get the e-fits, and why did Operation Grange wait 2 years until October 2013 before releasing them for Crimewatch? Part of the answer may be they were hoping something more promising would turn up, but by Crimewatch even a very low-grade source was considered to be worth a roll of the dice.