AMITIAE - Tuesday 31 December 2013

Cassandra: Identifying Persons and Locations from Eye Reflections in a Photograph

apple and chopsticks


By Graham K. Rogers


An online article by James Trew on Engadget outlines some new technology with the potential for identifying criminals using eyes in photographs. The article mentions the TV program, CSI, but this was also one of the technical gems of the 1982 movie Blade Runner, when Deckard (Harrison Ford) - using voice commands - was able to crop and enhance an image to produce a likeness of a subject from a reflection: now Life mirrors Art.

The Engadget item reports on research at the University of Glasgow carried out by Rob Jenkins and Christie Kerr of the University's Psychology Department. The journal article is available online: "Identifiable Images of Bystanders Extracted from Corneal Reflections". They examined scenarios in which, for example, cameras are seized during investigations and that the use of such reconstruction could help identify associates or bystanders. The Glasgow research also looked at the way this could be used with face-recognition techniques: software; and identifications by persons.

Jenkins and Kerr used a Hassleblad H2D-39. This produces high quality images with its 39 million pixel sensor. Hasselblad cameras are not cheap, nor are there many security cameras made by the Swedish company. The H2-series has now been superseded by the H5-series.

Earlier this year, I took a photograph of a friend that had such good focus (I guess I was lucky) that I was able to enlarge his eye and see myself in the image. I later had that printed as a 20" x 30" poster. That was taken with a Nikon D7000 camera and a 24mm lens. I re-examined some of the images I took this week on a trip to Mahachai out of Bangkok. I was using the same Nikon camera, but took several shots with a new 85mm lens I had just bought.

Eyes I also tried with the iPhone 5s and its FaceTime camera (1.2MP photos - 1280 by 960).

Although the quality (picture right - 61 x 61 pixels) of the iPhone output is nowhere near that of the Nikon (and certainly not the Hasselblad), there are elements within the image that might be identifiable in terms of location were the technology to be improved.

However, with my blue eyes, information seems to have less definition than when photographs of those with brown eyes (see above and below) are used.

Looking at a number of the best closeup shots, I enlarged them with the loupe (a magnifying tool) in Aperture and took screen shots. I set the loupe at 400% magnification. The results were reasonable for this DSLR camera: it is feasible that even without the far higher high quality output of the Hasselblad, there may be some identification possible.

Of course, most of the images show me - as photographer - but there is also some other information (figures and locations), depending on where the subject is looking.

Me in image center; bystanders and station facade to the right

The images here (which can be seen online) are cropped and slightly sharpened (1%), but the potential of identifying people (or locations) with photographs from even a relatively inexpensive DSLR camera is clear.

Me right of center; bystander to the left; station and train behind

Busy street environment

I know, I know, you have nothing to fear if you have done nothing wrong; but if you believe that, you do not deserve security or privacy.

Graham K. Rogers teaches at the Faculty of Engineering, Mahidol University in Thailand where he is also Assistant Dean. He wrote in the Bangkok Post, Database supplement on IT subjects. For the last seven years of Database he wrote a column on Apple and Macs.



Made on Mac

For further information, e-mail to

information Tag information Tag

Back to eXtensions
Back to Home Page