Frequency responses of half-pel filters – The ryg blog
fgiesen.wordpress.com
fgiesen.wordpress.com
I'm a 1D DSP guy so I could be totally off here.
But iterating an FIR is basically this
_ _
+ -> |_ G _| -------> y
^ |
|______________|
With initial conditions of G being the pixel value, right?I suspect the poles of that system are the eigenvalues mentioned in the post, but I don't really want to do the math. Then you can look at the effects of coefficient quantization as a pole/zero loci problem.
I think you should be able to guarantee stability (at the cost of phase distortion, and maybe overall gain?) by reflecting some of the zeros of G into the unit circle so they don't form unstable poles, but if they wind up being close to the edge then you still have to worry about quantization effects.
But it's been a long time since I've done image processing, so I don't know how relevant that is.
https://caseymuratori.com/blog_0035 https://caseymuratori.com/blog_0036