Examining the
HDTV snapshot, the "grain" doesn't look like grain.
R-G-B separations, compared among themselves, show the "grain pixels" varying lighter and darker similarly across all three, regardless of the area color. That is not grain, which varies independently in each separation. It's just noise. How did that noise get in? Beats me, but it's got to go.
Reducing that noise in
R-G-B separations is pointless. Each separation would be processed identically and it wouldn't be different than processing the full picture just once. However, it can be done in
H-S-L separations. Why use separations at all? Because it allows one to surgically adjust only those areas that need it. Not touching other areas that don't need it allows the picture to retain it's character.
Now, forget adjusting
Hue -- try that and it's
60's psychedelic time.
Saturation can be smoothed, which I did, to unify pixel intensities that are a jumble even in solid color (you have to see that separation to believe it).
Luminance, however, is where the real work is done. In a paint program, I use "JPEG artifact removal" on
low or
normal, a couple of times to reduce it's noise. Then a low-percent "Unsharp-mask" is applied to put the crispness back into any softened edges.
When the
H-S-L separations are rejoined. the picture is a little weak after all that processing. In "Hue/Saturation/Llightness", a slight increase in "saturation" and slight decrease in "lightness" restores the overall look to that of the original ... except for the noise.
This processing may be useful, even to those who hate the industry's
DNR madness.
Before and after at 200% magnification: