Willie Smits

That is what I did today. But not the usual way, the way I like it best, looking straight into their understanding eyes. No today, May 4th, 2011, my loyal friend Jan Geerdink, formerly with the Dutch police for fingerprints, had arranged for me and Jean Kern to visit Logica, a biometrics company in Rotterdam where we were awaited by Ronald Hack and Yosuf, a student that had taken the subject offered by Logica for a student to test various facial recognition software packages for use with orangutans. So we looked at some pictures and how the software saw those pictures, in terms of a network of points distributed over the face that all together with some sophisticated algorythms represent a value that then can be used to identify that orangutan.

Test application Logica

Yosuf had tested four software packages and soon found out that the way those programs work are much better at recognizing people (more than 90% accuracy scanning public in a football stadium) than orangutans of which they were only able to recognize 3-5 out of 761! The problem is that the eyes of orangutans do not have the white in the eye, the contrast needed by the program to identify the location of the eyes from where the rest of the points in the face are then mapped. Orangutans have almost all brown eyes with the iris color extending far into what is white in our human eyes.

But why would I be sitting there watching and listening to Yosuf presenting his study results to us? Well actually for a number of very good reasons. I have always hoped to find a way to recognize orangutans after their release. Since most of the orangutans that we release are still rather young, they tend to look very different a few years after their release in the wild. Their facial color is much darker, some males will have developed cheek pads, their hair will be much longer and just as with humans some expression representing their individual character becomes like engraved in their faces. I learned that the human face from year to year can change some 5% so with orangutans it is no wonder I have sometimes problem recognizing friends after encountering them again after several years of them living free in the forest.

Orangudome

But we can do more with the facial recognition software. I want to use it in the Orangudome in Yogyakarta, so we can monitor what orangutan is where and can arrange their habitat in such a way that conflicts are avoided through a system of hydraulics that can either connect or disconnect different parts of the seven intermingled 3D worlds that will make up the inner space of the Orangudome. That would mean it would have to work in real time… But it would also be great when we could use undercover recordings, made with hidden cameras, to check what orangutans may have been smuggled or stolen. We also may be able to let the software of Prof. Paul Ekman work better when we know the individual orangutans whose emotions we try to identify from the combination of facial muscles contracting.

So lots of reasons to be interested in the facial recognition software with which Logica works. But with a score of 3 out of a 761 pictures… is there any hope??? We had an interesting discussion on how to deal with the problems of locating the eyes. With humans this is the most conspicuous feature of a human face, the location of the eyes. But the face of orangutans contrasts less with the extended irises of similar color, so sometimes the software would place cross marks on the cheeks to indicate where it calculated eyes should be or the nose would sometimes be seen as location of the eyes. When the eyes were located manually in the picture than the system would work better and the incidence of orangutan individuals correctly identified from pictures would increase much, but still not very good.

I suggested to Logica that they start with looking at the lighter part of the head and the rounded form of the skull in combination with the better lighted part of the protruding mouth and based upon those two contrast rich regions let the software look for the correct position of the eyes. Then to look at the angle between the eyes and the nostrils, which I found to be very consistent and different between orangutans. Ideally one should then have the orangutan looking straight into the camera. But when we take a flash picture we can locate the reflecting pupils better and at the same time locate the angle under which the picture was taken so we can mathematically correct the angle between the eyes and nostrils to represent the straight looking position.

Sounds complicated? Not really. Yosuf got it and will make some tests with the system I proposed. Another possibility is to use a scanner, with some kind of beam that will reflect from the face, perhaps infrared, or another wavelength of light, or the body scanner frequency now becoming standard for checkpoints at airports. I promised to have a look at the thesis of Yosuf and then we left for The Hague to meet the Indonesian Embassy.

Another important step towards helping orangutans assisted by good willing people in The Netherlands using modern technologies! Now not to identify hooligans or terrorists but protect and help innocent orangutans.

Willie Smits, Amsterdam, 4-5-2011

Willie at Logica

From left to right: Ronald Hack (Logica), Yosuf Haydary (University of Applied Sciences of Rotterdam), Willie Smits, Jan Geerdink (former Dutch police). Picture taken by Jean Kern.