Stochastic Resonance in Vision

Stochastic resonance described in Mind Hacks:

…adding noise to a signal raises the maximum possible combined signal level. Counterintuitively, this means that adding the right amount of noise to a weak signal can raise it above the threshold for detection and make it easier to detect and not less so.

Link to demo: [Link]

without noise
with less noise

with noise
with more noise

6 Responses to “Stochastic Resonance in Vision”

  1. Why are the little white dots in the “with more noise” considered to be noise? On the contrary, it would appear that they are signal. Or does “noise” merely mean some sort of vaguely conceived randomness? True randomness is very difficult to generate. The dots in the “more noise” picture are not at all randomly placed. They are positioned to make a face. Or was the point supposed to be that what might seem like noise is really signal?

  2. Whit Schonbein says:

    I think what’s missing from the description is that (i) the original signal contains information regarding whether each pixel should be on or off and (ii) the signal is not always strong enough to cause pixels to turn ‘on’. So, for each pixel, imagine the original signal has a 0 if the pixel should be off, and a .90 if the pixel should be on, where the threshold for actually turning on a pixel is 1.0. Adding some (white or pink, i guess) noise to the mix has a greater chance of pushing the ‘on’ portion of the the signal above the threshold than pushing an ‘off’ portion portion above the threshold, which is why the white dots appear to selectively target the relevant parts of the image when noise is added.

  3. Pete Mandik says:

    Whit, I think that sounds right. The way I think of the visual cases is you start with a pretty clear picture and then mask an increasing number of portions of it with black pixels until you have something just below threashold for recognition. Adding certain amounts of noise (snow) to the masked portions bumps the brain’s detectors up to a “yes” response.

    An auditiory version might take a recording of an uttered word and turn down the volume until just below recognition threashold and then mix in the sound of a washing machine to facilitate recognition.

    A philosophical version might involve randomly inserting the word “fnord” in all of my arguments to make them more convincing.

  4. Does the phenomenon where looking through eyelashes makes noisy pictures clearer have anything to do with this?

  5. Pete Mandik says:

    It might, Tanasije, but I’m not sure. Makes sense.

  6. Eric Thomson says:

    Tanasije: I think not. That more likely implements a low-pass filter that smooths over the pixellation in the noisy pictures and the resulting shading effects bring out more recognizable patterns. We could test this by putting an overhead slide covered with scotch tape over a pixilated image and seeing if, with eyes wide open, we see more than when the low-pass filter is not present. We could test your alternative theory by adding noise to such an image and seeing if we better recognize it. My hunch is no, since it isn’t a matter of things being subthreshold, but rather being suprathreshold but noisy. I don’t think you are adding noise by squinting, but just low-pass filtering.

    But I’m not sure. This could probably be looked at easily in Photoshop with various filter and noise-adding functions.