23
$\begingroup$

Is there a generically accepted method of removing motion blur from an image. For a simple case, we can assume that motion happens in a straight line. I assume that it's a two part process composed of motion estimation and then deconvolution, but how is it actually done?

$\endgroup$
12
$\begingroup$

Yes, deconvolution. This page describes a number of deconvolution methods and methods for estimating the point spread function:

They say the deconvolution literature is "extremely extensive". They choose Lucy-Richardson algorithm for deconvolution and develop their own motion estimation algorithm for determining the point spread function.

before after

$\endgroup$
9
$\begingroup$

Removing motion blur, with the exception of blur from either point source objects, or near point source objects, can be quite difficult. The general steps are as follows. Note, some of this information comes from a paper entitled "PSF estimation using Sharp Edge Prediction".

  1. Find the equivalent PSF for the motion blur. This may vary in different regions of the FPA, depending on how wide the lens is, and what kind of motion it is subject to. The easiest way is to find a point source object, and find it's PSF. This can be accomplished if the background is uniform, especially if it is uniformly dark, such as stars. A more complex method could involve looking for edges, and seeing the effective PSF at that edge. The edge will give you a one dimensional PSF, but can be used to find a complete PSF.
  2. Using the PSF, deconvolve the image. As Endolith previously mentioned, there is a plethora of deconvolution algorithms, choose one to meet your needs.
$\endgroup$
0
$\begingroup$

Just an idea: maybe one should study some HDR imaging for this task. There is a paper about radiometric camera calibration for composing correct HDR images. It is shown that there is a huge difference between the images resulting by applying synthetic motion blur on LDR and HDR images.

There is a page for this: http://ict.debevec.org/~debevec/Research/HDR/

Motion blur on LDR and HDR images:

LDR HDR

$\endgroup$
  • 3
    $\begingroup$ Your answer is not related to the question. OP specifically asks for removal of motion blur. Your answer suggests that applying motion blur to HDR (intensity corrected) gives more realistic visual appearance, which makes it a technique in computer graphics and can't be used to remove motion blur from a photograph. $\endgroup$ – rwong Sep 11 '11 at 23:06
  • $\begingroup$ Sorry for not being clear. I suggest studying HDR imaging because removing motion blur from a real image (having the saturation effect caused by the low intensity range the camera can encode) can involve deeper inspection. Having a blur remover method for the first image will probably not make the job for the second image. So my notion is relevant. $\endgroup$ – Bálint Fodor Sep 12 '11 at 10:40
  • $\begingroup$ When performing deblurring, it is understood that (1) saturation should be prevented or else it would make deblurring impossible (loss of information at capture time), (2) the pixel intensities be normalized to a linearly additive response curve. Deblurring cannot be done without these two prerequisites. $\endgroup$ – rwong Sep 12 '11 at 17:15
  • $\begingroup$ Hello @BálintFodor, in light of the comments, could you perhaps clarify and explain how your answer relates to motion deblurring? Your comment below rwong's isn't very illuminating for those not well versed in the subject and I'd really be interested in knowing more. Thanks :) $\endgroup$ – Lorem Ipsum Nov 19 '11 at 0:23
  • 1
    $\begingroup$ Hello @BálintFodor, after reading the paper, I am ready to undo my downvote. However, the site voting mechanism requires you to add more information to your question before I can change my vote (it's a mechanism built into StackExchange). $\endgroup$ – rwong Nov 20 '11 at 23:44

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.