If you show a picture, really quickly, faster than the "refresh rate" of the eyes, wouldn't it just be blended or motion blurred into a surrounding frame?
Suggestion for methodology: First show neutral grey. The flash image for deltat. Then flash negative image for deltat. Then neutral grey again.
This ensures the average over time is grey.
I think for many of these studies they have to go even further because it turns out that people can remember the target images and "look" at them (in their mind's eye) for longer than the allotted time, even if sandwiched between neutral frames. So the target images have to be followed by random noise images that have the same frequency statistics as the target images, which interferes with your brain's recall mechanism and prevents you from "cheating."
Lots of interesting work on this by the military. They show rapid-fire photos of satellite imagery to analysts who are looking for interesting features. They monitor the brain activity and look for responses when features appear. Results show that the analysts' brains detect things before they're consciously aware of it.
They then tag the images that registered a response and look at them in more detail. Apparently it's fairly effective.
Presumably, those analysts still need to be trained in the normal way first so that they know what features they're looking for.
In addition, you could probably only use people for short durations before you have to allow them to rest, otherwise you run the risk of exhausting people and missing things (c.f Air Traffic controllers).
Apart from that, this kind of reminds of the 'Focused' people from one of Vernor Vinge's novels [1].
The process of transducing light into neuronal spikes adds significant temporal smoothing, so an image briefly flashed on the retina will be lengthened in time somewhat before it even makes it to the brain. Need to look up exact numbers here, but this isn't all that surprising.
What is quite surprising is that rod photoreceptors are sensitive to single incident photons. That's truly unbelievable.
Back in the mid-90s, when I was doing digital editing, it got so I could easily spot a single bad field (often just a portion of one television scanline) appearing for 1/60th of a second (or about 16 milliseconds).
It wasn't quite "seeing" though, I was just suddenly aware and when I jogged back to the bad frame, there it would be.
In an Italian TV show, SuperBrain - Rai1, I've recently seen this guy, Alessandro Gallippi, who is able to recognize a movie from a single frame embedded within another movie.
Pretty impressive!
Here's the TV show [1], an example is at 1:12:07. Of course I can't know if that's fake, but I have no reason to think so...
This reminded me of Ayumu the chimp [1]. The difference between being able to perceive the overall image like a zebra, and the specific details like the position of nine numbers seems interesting to me.
This reminds me of those courses that supposedly train you to read books at insane speeds by relaxing and taking in the pages like a photo. Is there anybody that actually gets results with this approach?
Those one frame flashes are easy to see though (which makes me wonder where the previous result of 100ms treshold for image recognition came from). Probably coincidence resulting from similar frame rates, but I think movie projectors were running 72 hz shutter showing each frame 3 times to provide low flicker image. That amounts to 14~ ms per showing which is close to what they tested here. Of course part of the time the shutter is blocking (50%?) so the actual image might be projected only half of the time.
This is not fast, as 13ms is only ~75Hz. I'm not sure why they claim that the results are surprising. What's surprising to me is that they had a test setup that supported only 75Hz.
As for Carmack, I belive he has been onboard for even 200Hz displays for qutie some time already.
