I need to estimate the “true resolution” (don’t have a better word for it),
to identify
- poor optics (blurry, foggy, unsharp)
- poor sensor (i.e. noise. Sensor too small, bad tuning etc.)
- post “artifical” resizing
I found there are some very academic “direct” approaches for that.
Still… “I had an idea” - and I wonder if that “poor man’s approach”
might deliver acceptable results.
Take the orig image
- shrink it with a detail-preserving alg (Lanczos etc.).
- then resize back to orig size
- build pixel-wise diff
- if more then a special amount of pixels differs in more than some value:
Orig image has more resolution than the tested shrink value. Else less. - Do interval nesting on the resize values
Will this do? Or complete nonsense.
Maybe too slow comparing to other algs or
too much dependent on some invariant image characteristics, etc.
TIA 