- by SRS1, Master Analyst
- 1/20/2014 6:47:57 PM
This all sounds facinating and something that will be utilized in the future. However at this time I don't think 39 megapixel digital cameras are cheap enough to be used in large numbers. You never know though because 3 years ago no one would believe that we would have 5mp (or higher) cameras on most new phones.
- by Michael Steinhart, Blogger
- 1/9/2014 6:02:47 PM
That's certainly a worthwhile avenue of research, Beth. And the fact is, resolutions are getting higher and image sensors are getting cheaper, so the possibility isn't very farfetched.
- 1/8/2014 6:13:01 PM
@Michael, your points are interesting, but I think we're getting ahead of ourselves to leap from research lab to stills captured from a video surveillance camera. I'm more envisioning this work to play into investigate work in kidnappings and child pornography. For example, the kidnapper sends a photo to prove he or she has the victim and the victim is still alive. Investigators mine the victim's eyes to see if they can discern who any of the kidnappers might be from the reflections in the eyes. Likewise, investigators mine the eyes of children in pornographic photos to see if they hold any reflections of the picture-taker. I would imagine investigators of these sorts of crimes would be interested in hearing more about the study.
Or maybe I've been watching too many tech-savvy crime dramas!
- 1/8/2014 6:07:10 PM
@kq4ym -- to be sure, criminal investigators looking for potential perps would not find this type of effort fruitful for now. But the promise is interesting to me. Here's more detail from the research, regarding the quality issue:
For the current study, we used a 39 megapixel digital camera to take passport-style photographs of volunteer models. In separate exposures, these volunteers served as subjects, when they were direct subjects of the photographs, and as bystanders, when they were visible only indirectly via the subject's corneal reflection. Pilot work determined that image area for reflected bystander faces was smaller than for subject faces by a factor of around 30,000. The quality of the bystander images is thus poor, despite the high pixel count of the source photographs. To establish whether bystander faces could be identified from the extracted images, we presented them in Experiment 1 as stimuli in a face matching task –. To assess effects of familiarity on match performance, we compared observers who were familiaror unfamiliar with the faces concerned. In Experiment 2, we assessed spontaneous recognition of the extracted images.
- by Michael Steinhart, Blogger
- 1/8/2014 4:38:07 PM
I agree with kq4ym. The resolution at which you could blow up a tiny segment of a photo and still have enough visual data to work with that you can analyze it would have to be professional-quality. And what about red-eye?
Are the surveillance cameras currently in use in public areas this finely tuned?
- by kq4ym, Data Doctor
- 1/8/2014 3:44:46 PM
From what I've read in press accounts of the relected image possibilities, the camera needs to take a very high pixel image and needs quite a bit of very good lighting to make this possible at this time. While it might make an interesting story as someone takes secret pictures of us all and places us into that big data bank, it's probably not going to be quite as useful as existing pictures of most all of us already existing in driver's license photos and our voluntarily uploaded stuff to Google and Facebook among hundreds of other places we're all found.