I have been struggling with the LR blind deconvolution for a little while...
Does anyone have a working code snippet C/C++/C# that shows the initialization of the kernel, and which kernel size to use?
The Intel documentation is not very informative on this I think.
My images are 2D and typically 1600x1200 pixels. It would be great to be able to run RGB deconvolution - but mono would be some start...
Any hints or comments are highly appreciated
I looked at the kernel size discussion, and came a little further - now the program runs without crashing, and without bad error codes :-)
Unfortunately the resulting image has some artifacts - like horizontal lines in the image, and some repeated 'error-pixels' (Would be easier to explain if I could attach an image...).
Some parameter questions:
(1) I briefly looked into the original 1972 Richardson paper, and found that the number of iterations seems to be 10 in the examples there.
(2) What is the purpose of the thresholding parameter?
(3) Like others I have a tough time understanding the kernel, its size and sane values from the start. The original paper uses a 2x2 kernel?
(4) From another discussing in this forum, I read that the ROI needs to be square - or maybe it is the kernel? It is not very clear.
A running sample would be great - it is an interesting topic!