DSP Methods for Blur Reduction
Conclusion
It is no revelation that image reconstruction is important and effective.
However, it appears as though there is, of yet, no ideal method for reducing
blurring in images.
The Wiener filter works, of course, and, in our findings, works quite well.
However, it is rarely the case for the response function of the degrading
system to be known. It could be guessed at, but that is an undesirable burden;
the number of possibilities just for blurring functions is huge.
The Iterative Blind Deconvolution algorithm works very well in some cases;
in others, it fails to converge and the results must be observed at each
iteration to choose the best image. Also, some guesswork is involved here
regarding inputs to the program. We found several interesting features:
- The result is very sensitive to the size of the PSF; in general, it is
desirable to use the smallest possible value
- The result is insensitive to the actual estimate of the PSF
- The result is sometimes sensitive to the estimate of the image; in some
cases, using the blurred image itself as an estimate worked well, in others,
using a white box worked well.
- In general, if a good result will be produced, it will be within 10-20
iterations
In all, this appears to be a promising technology, but still more work needs
to be done to improve some aspects of its behavior, and to get a better idea
of how to choose all the parameters.
The Minimized Constraints method, in theory, solves the problem of guessing
parameters, as well as the convergence problem, in the IBD algorithm. However,
our testing does not show it to be particularly effective.