11
$\begingroup$

Everybody knows the effect of pixelated objects (e.g. faces) on TV. Is there a way - and which mathematical method lies behind it - to un-pixelate the region? Beware: I am not talking about smoothing it out (unblur) image by image - but about using the information of a whole sequence of pixelated pictures where the object which is pixelated naturally changes the location within the overall pixelated region. I guess it must be some kind of "un-averaging over time" (e.g. differentiation over time comes into mind).

Has anybody tackled this problem yet? Are there references, demos, software...?

Thank you.

Improve this question
$\endgroup$
8
  • 1
    $\begingroup$ Offtopic ? $\endgroup$
    – Qfwfq
    Jan 10, 2011 at 15:03
  • 5
    $\begingroup$ @unknowngoogle: No, I don't think so because the underlying math should not be trivial. $\endgroup$
    – vonjd
    Jan 10, 2011 at 15:04
  • 1
    $\begingroup$ Sounds extremely hard to do. To do that, you should first obtain some information on how the pixelated object moves in successive frame. And that is a difficult problem in itself, even on plain un-pixelated images. $\endgroup$
    – Federico Poloni
    Jan 10, 2011 at 15:20
  • 6
    $\begingroup$ @unknowngogle: Yes, it does sound like a question in applied math, doesn't it? I don't think it makes it not research-level or unsuitable for MO, though. $\endgroup$
    – Thierry Zell
    Jan 10, 2011 at 16:34
  • 1
    $\begingroup$ You might also try the astronomy literature --- all the time they deal with point sources being pixelated. It's been a long time since I was an astronomer, so I don't have any references for you, sadly. $\endgroup$
    – Scott Morrison
    Jan 11, 2011 at 21:47

1 Answer 1

Reset to default
14
$\begingroup$

This is a problem that has been addressed by the computer vision community in the past few years. You may want to look into upsampling and super-resolution techniques as well as vectorization methods. Here is a recent paper on the topic in ICCV 2009 which gives a good summary of the recent efforts in this field: http://www.wisdom.weizmann.ac.il/~vision/SingleImageSR.html

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ This is extremely helpful - now I have the keywords for finding out more - obviously there is some high-level research going here - Thank you! $\endgroup$
    – vonjd
    Jan 10, 2011 at 15:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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