Quantization vs. Subsampling

Igor R. wrote:
> I prefer more DCT compression then Chroma subcoding.

And here we can come close!
I do also think that we can avoid all that sampling issues,
BUT only after introducing SEPARATE quality selectors
for color and luminance!
So I would go along with you and prefer only 1x1 (no)
subsampling, IF we would replace the subsampling selection
by a separate color quantization selector. I had this idea long
ago, and think I will investigate that further someday.

But today you still have that different subsampling settings
around, so for now you have to deal with that anyway.

An odd thing with that non-1x1 sampling modes is that you get
different results with different decoders, depending on the particular
variant of upsampling used in the decoder (there are 2 different
variants offered in the IJG library - simple pixel replication and bilinear
interpolation, and I use yet another - better of course ;> - method
in my modified library for Jpegcrop).

The luminance and color already have separate quantization matrices
in JPEG, so you can easily handle the color reduction by quantization
adaption (in DCT domain) rather than pixel subsampling (in spatial
domain). But to get a similar quality effect, you need a coarser color
quantization adaption than luminance (-> separate quality scale factor).

My new resampling scheme is even equivalent to a corresponding
quantization adaption! But adapting the whole quantization process,
not just scaling with a factor. That's why I still want to further
investigate that method for application in the future...

Regards
Guido