Of course that is an entirely subjective question. However, without showing lots of pictures- I think the "answer" depends on the data itself.
To be a "good" image it generally has a self-consistency in chrominance, luminance, and graininess. The point of the question is whether a
smoothing process (noise reduction) results in pixel-to-pixel relationships in low S/N that relate to their high
S/N counterparts. The better the transition from one to the other, usually the better the visual appearance of the image's internal
self-consistency.
Perhaps also important is that high signal areas can have many contiguous pixels of equal values (or values that vary as a smooth gradient).
Low SNR areas (or background-like values) *naturally* show more variation. If variation in the low signal areas is minimized by smoothing-
and making many contiguous pixels have nearly equal values (or gradients)- this effectively promotes the background to be more high SNR like.
This means the background visually becomes a "thing" and not a natural background with variation as expected. This is the "waxy" bit that I think is often described.
This is why leaving some variation in background is good- it really is the form of the variation that is the key. We seek to find patterns in the noise. If you don't have
"good" variation- our brains say "I see the pattern (the thing)" where we really want to ignore it compared to other bits.
So it wasn't a thousand words- but the words above might be useful when looking at an example picture.
-adam
