8

Is there any strategy to detect the region of an image that is sharp and in-focus, in order to separate the out-of-focus background? I've tried using edge detection methods but without any success.

In other words, is it possible to determine if an area of an image is blurred (out of focus) or not?

Nathan Reed
  • 25,002
  • 2
  • 68
  • 107
p2or
  • 596
  • 1
  • 7
  • 15
  • 1
    Do you mean that you want to detect areas of an image that are sharp and in-focus, versus blurred and out-of-focus? – Nathan Reed Oct 07 '15 at 19:46
  • 1
    How did you used edge detection algorithms? Not sure it works and can't test now (so no answer for now :P), but just off the top of my head, you can divide your image in subregions and measure the value of gradients. Based on a threshold you can then decide whether that subregion is "in-focus" or not. Then you can re-use the edge info to refine your first approximation I described before. It makes sense to me know, but it may very well be a brain-fart being it late here :) I'll give it a try tomorrow if I have a minute to. – cifz Oct 07 '15 at 21:13
  • @cifz - Thanks! Great idea to measure the gradiants, sounds like a simple and fast approach. But this will only work for a rough detection, right? – p2or Oct 08 '15 at 10:24
  • @NathanReed Right. Sorry, not a native :) Do you think I should rephrase the question? – p2or Oct 08 '15 at 10:28
  • @poor Yes, I think it would be helpful to rephrase and edit the title to make it more clear. – Nathan Reed Oct 08 '15 at 17:10

1 Answers1

11

it is quite easy to measure the local max frequency in an image (at least as a low resolution mask, with some regularization). Several papers of the MIT graphics group have been around detecting and processing from this kind of clue, with regular or coded aperture cameras. e.g. Defocus Magnification and Image and Depth from a Conventional Camera with a Coded Aperture.

Fabrice NEYRET
  • 1,266
  • 8
  • 14
  • Many thanks Fabrice! If possible, can you elaborate local max frequency? – p2or Oct 08 '15 at 10:44
  • I don't remember how these papers were estimating the native focus before refocusing. But I could think of various solutions: poor-man wavelets by calculating the FFT in a grid of subimages (i.e. a grid of windowed FFT), estimating local autocorrelation, poor-man FFT by convolving with a set of Differential of Gaussians (to detect ranges of focus), etc. But this is kind-of instant-hacking, the best is to read what the experts did for real :-) – Fabrice NEYRET Oct 08 '15 at 12:58
  • Any work done on single image, not with Coded Aperture? Thank You. – Royi Oct 19 '15 at 15:08