After seeing this question I couldn't help but wonder if it's possible to take some (possibly very large) number of photos of a very far away object (several kilometers) distorted by atmospheric effects like air temperature difference and humidity, and using some algorythm pick only the least distorted portions of each image, and touch up the rest. Is it a thing or is it just never done in practice?

I understand this would be very hard with moving objects, so I'm only asking about stationary objects.


The technique you are asking about was mentioned in one of the answers to that question, lucky imaging. This technique is used in astrophotography to resolve objects that otherwise are unresolvable by any single image, to cancel out atmospheric effects.

In principle a similar technique could be used for terrestrial subjects. However, it's only applicable for stationary subjects. Moving people, animals, cars, etc., cannot be better resolved with this method.

  • I guess I should clarify I'm only asking about stationary objects then. Thanks for the info. – user1306322 Oct 18 '16 at 13:35
  • 1
    i would imagine that various intelligence agencies do employ computational photography techniques, but they don't exactly distribute their software as open source... – ths Oct 18 '16 at 13:37
  • @ths you mean "Enhance" ? :-) . On the serious side, until relatively recently, rather a lot of intel analysis depended on so-called experts who claimed their experience allows them to distinguish a random shadow from a secret Soviet missile-silo door. There was a lot of wishful thinking involved. – Carl Witthoft Oct 19 '16 at 11:29

What you need to look up is the term "image restoration". It is an area of active research.

The Wikipedia article is not very informative, though .

If the distortion function is known (e.g. when you know the camera has been moved only horizontally) you get better results than in scenarios where the distortion function is unknown. This is called "blind restoration".

Nevertheless, many such methods exist, for photos and videos. What they can do is illustrated with a screenshot from slide of a MOOC I am currently taking:

blind restoration

The reference at the bottom of the picture describes Bayesian methods, Total Variation (TV) and Simulataneous Autoregressive (SAR). Just to explain the abbreviations. Actually I don't know how these algorithms work in detail.

SNR = Signal to Noise Ratio

Another related technique that does what you want (for static objects) is called super-resolution imaging.

  • Adobe Photoshop has a related feature called "Shake Reduction" tough my guess is that this would not work on subjects with minor random blurring such as you get with distant objects. helpx.adobe.com/photoshop/using/… – Jeremy Oct 18 '16 at 20:28

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.