Slashdot Mirror
CSI Style Zoom Sees Faces Reflected In Subjects' Eyes
mikejuk writes "A recent paper by Dr Rob Jenkins of the Department of Psychology at York University (UK) has managed to prove that you can get useful images of faces from the reflections in eyes. It really is as simple as zooming in. The catch is that the experiments were done with a 39 mega pixel camera — even so the actual final images were low resolution. In the experiment a number of people were photographed with a 'bystander' in a position so that a reflection of their face would be captured in the eye. The resulting extracted image of the reflection in the eye was only 27x36 and then rescaled using bicubic interpolation to 400x240 or bigger and enhanced using standard PhotoShop operations to normalize the contrast and brightness. Test subjects were able to match faces using the low resolution images but the important result was that if the subject knew the person in the photo then recognition went up to 90% with false positives down at 10%. So the next time you appear in a photo consider the fact that a simple procedure might reveal who you are with."
Coming to NSA-approved monitoring sites like Facebook: automatic detection of people you are with, even if they're not "in' the photograph!
Is it just my observation, or are there way too many stupid people in the world?
Gratuitous enhance!
As a photographer, I have to clean up the eyes and all the other reflective materials from my images, if any, or find myself in sweet spots where it doesn't happens.
It's also common knowledge that you can actually learn about the photographer's lighting setup and materials by looking at the shadows AND light reflection in the eyes.
There are also nice non-shopped photographies with that reflectiveness property being applied to produce interesting results, with inverted sceneries being shown in the eyes. Take the picture close enough and you won't have 50 pixels, but 500 of reflection. Take it with a high resolution camera and you'll get 1000-2000 pixels worth of goodness.
So there's a direct correlation between a pixel count and the number of pixels for a detail.
In other words, it's known that you get good reflectivity in eyes and it's known that using high resolution cameras you can get good details from eyes, inferring you can tell who's on the other side. Well ... yep. Next thing they'll figure out is by having a high sensitivity camera you can get cleaner pixels of the people themselves in the reflectivity, as the lighting is seldom as adequate and the contrast is much less prevalent in these reflections, so by having much cleaner 32 native bits per color, you could technically have a "perfect" rendition of the user, especially at high F
Surprise me next time :) I'd even say that if you could have a consistent reflective model of a person's anatomy and a map of its clothings, you could potentially infer a person's surroundings, as every material is slightly reflective / refractive too. But we're really not there yet. :)
A recent paper by Dr Rob Jenkins of the Department of Psychology at York University (UK) has managed to prove that you can get useful images of faces from the reflections in eyes.
Didn't anyone think this couldn't be done?
Are you sure he doesn't work in the Department of Bleedin' Obvious?
systemd is Roko's Basilisk.
These guys at Columbia did something similar years ago, but used eyeball reflections and a cornea model to figure out what the person was looking at: http://www.cs.columbia.edu/CAVE/projects/world_eye/
Anybody with a recent DSLR can test that this can be done. I recently took a portrait if myself in my cat's eye.
The trick to this though is that you need a DSLR with fairly high resolution, a good sharp lens, and have the photo be a closeup of the subject. None of which are features of the vast majority of security camera footage.
If somebody was willing to spend amounts in the range of $1000 per camera, yes, this might be a possibility. Provided the person stared right into the camera while standing at a meter or so from it.
Also, it'll probably stay this way. There are limitations to the useful resolution that can be achieved, so it's not possible to simply put a 1000 MP sensor into a security camera and suddenly be able to perform the tricks shown in CSI.
... the technology to convert unfocussed VGA resolution photographs and convert them into focused 39 megapixel photographs. That is why this technique is so useful on CSI.
Thank you for bringing up such quality content. You know, 39 megapixel pictures are only usable if you have high end glass on your camera, which only rich camera enthusiasts can afford. That excludes forever smartphone cameras (yeah, even those whatever megapixels Nokias), security cameras, small digital cameras, which would actually render useless 99% of all cameras on earth for that usage. So you only need to have one of these rich guys take a picture of a crime scene, yeah, only that...
Stupidity is the root of all evil.
The pixel count increases as it's a selling point. The optics on most things like phones seem to stay the same utter crap, barely enough to saturate the ccd in full sunlight, leaving most of those megapixels as random noise. Leaving any CSI style zoom as something achievable in photos specifically taken with the proper equipment and setup to allow CSI style zoom but not in the junk quality images they pretend to enhance in CSI.