motionblur compensation

From: grxzmk (klap_at_gmx.org)
Date: 01/27/05 

Date: 27 Jan 2005 08:12:28 -0800

Hi,

I try to restore motion-blurred images, where the original image is
rich in contrast and sharp edges. For simplification, the motion is in
x-direction, so lines are independend and the problem can be
considered 1-Dimensional.
As model of blurring a line, I assume a convolution of the original
picture-lines and the pulse response vector (flashcurve, scaled by the
velocity). The relevant pulse response is only some (1-10) pixels
long.

First I have tried to recompute the original line by modeling the
convolution as linear system of equations and back substitution the
lower triangular matrix.
This seems theoretically perfect but practically there are
high-frequency periodic signals superposing the reconstructed line.
Second I did the same (inverse filtering) in frequency domain, with
similiar Results.

Now I read from the Wiener-filter and saw examples, where hardly
blurred pictures have been reconstructed nearly perfect. I am confused
because if there is no noise the Formula to compute the Wiener-Filter
seems to be the same as for Inverse-Filtering in frequency domain. Is
it, or do I misunderstand the way getting the wiener-filter?

Is it generally possible to reconstruct pictures with sharp edges
because of the high frequencies (caused by the sharp edges) and the
sampling theorem.
Does avoiding frequency domain help.

Thanks a lot,
Klaus