23

I have heard quotes which say "If its not good, delete it."

I go by that advice, but sometimes you capture some precious moment but you see that photo is slightly blurred because of camera shake. I don't have an IS lens, and it's too late to somehow do it over.

What best one can do in post processing to improve the quality of the picture?

Ideally, I'd like to use Gimp or Picassa for this.

Edit: I came across a video regarding this -
http://www.youtube.com/watch?v=xxjiQoTp864&feature=player_embedded

sat
  • 1,727
  • 3
  • 18
  • 28

3 Answers3

31

I used to think that blurring was one of those things that was impossible to recover from in post. Amazingly enough it's possible to take an image that is blurred beyond recognition:

and recover all the original detail if you know the exact blurring function:

So why isn't this done all the time? Well firstly you never know the exact blurring function so you can't create a perfect inverse filter, secondly if you have noise in the blurred image:

this will totally bias the outcome, as the inverse filter is unable to replicate it:

pseudo inverse filters such as the Wiener filter can cope much better with noise but you still get ringing artifacts like the following:

image (c) MathWorks, see http://uk.mathworks.com/help/images/examples/deblurring-images-using-a-wiener-filter.html for more detail

This is a bit of a digression, but it shows that deblurring is at least possible in principle. There are some very clever algorithms that outperform the Wiener filter by guessing what parts of the original image looked like, in order to estimate and reverse the blurrig function, based on the statistical likelihoods of various light patterns existing.

There are some Photoshop plugins that offer image deblurring using such advanced methods, you might want to take a look at the following (which offer free trial versions)

The results are never perfect but for shots that are irreplacable it's better than nothing!

clabacchio
  • 2,219
  • 3
  • 17
  • 31
Matt Grum
  • 118,892
  • 5
  • 274
  • 436
  • Wow - that's pretty impressive. I guess that underscores the importance of saving all those marginal shots (preferably in RAW). You never know when medical science will advance to the point where the patient can be saved. – D. Lambert Nov 08 '10 at 14:41
  • @D. Lambert: Always hold on. Yesterday I took a shot that was 8 stops underexposed on my K-5 and was able to recover what was essentially a black box into a quite usable image. – Joanne C Nov 08 '10 at 15:18
  • @Matt Grum -- This is just too good !! Its such an enlightening piece of knowledge and Information (to me at least !!) – sat Nov 08 '10 at 18:00
  • 1
    Stuff like this is still more theoretical than practical, but a lot of the stuff that's now standard in Photoshop (content aware scaling for example) first appeared in scientific journals. – Matt Grum Nov 08 '10 at 18:35
  • 4
    Also: If you have motion-detecting sensors in your camera (like the gyros/accelerometers in an iPhone, or the sensors used for image stabilization) that data could be used to aid the creation of the filter. Perhaps in the future the camera makers will start embedding that data in the images. http://research.microsoft.com/en-us/um/redmond/groups/ivm/imudeblurring/ – coneslayer Apr 27 '11 at 16:50
  • Sounds like a fun iPhone app to write. – Evan Krall Apr 28 '11 at 06:10
  • Hello @MattGrum, it seems that the pictures are not available anymore. Could you retrieve them? – clabacchio Nov 26 '14 at 10:09
  • @clabacchio done, for now, will need to rehost these in the long term... – Matt Grum Nov 26 '14 at 11:27
  • Actually I think that the Mathworks example is a bit misleading: what it means with "blurring function" is the function that remaps every pixel to its new value. And you never have this information available, so the result will never be that good. Not even close. – clabacchio Nov 26 '14 at 12:18
  • @clabacchio that's exactly what I say in the answer, the images demonstrate the theory, not the practice. – Matt Grum Nov 26 '14 at 15:08
  • Indeed, I didn't want to detract from the answer, I was just suggesting to stress the difference between the theory and practice, and the difference between a function that describes the cause of blurring (e.g. subject movement) from the actual effect on each pixel. – clabacchio Nov 26 '14 at 15:11
  • @clabacchio - As I read the answer, Matt answered all of the items you suggested. My impression from your comments were that you didn't actually read the answer, which, candidly I've made the same mistake at other times. – B Shaw Nov 27 '14 at 14:34
  • @BShaw I'm not saying he didn't answer (he did), just that there are slightly ambiguous definitions, like the "blurring function" specifically, that can be misleading if one doesn't know exactly what they refer to. – clabacchio Nov 27 '14 at 15:31
  • Would this not be feasible in eg nightshots that have stars in the background - you literally have a perfect example what the blur has done to an object you know should be circular and small.... – rackandboneman Jan 22 '19 at 20:04
2

If the photo is blurred due to camera shake, you really don't have many options. While lots of things can be fixed after the fact in post-processing (such as white balance or exposure), camera shake can't be undone. About the only option here is to apply a blur effect to more of the photo that might make it look like it was intentional.

Even without an IS lens, most newer DSLRs (and some of the high end point and shoot cameras) have great performance at higher ISO levels. If you find yourself in a situation where you can't get as fast of a shutter speed as you need, bump up the ISO. Sure, you might get a bit more noise in the photo, but noise can be addressed via software if you wish.

ahockley
  • 21,446
  • 6
  • 75
  • 147
2

You can try this software: http://smartdeblur.net/ It's able to detect blur kernels automatically.

Ivan Kochurkin
  • 121
  • 3