Chroma in Downsampled 2x1 JPEG?

Bill Tuthill wrote:
> When 50% downsampling digital camera images with 2x1 chroma subsampling,
> (for sending a CD to friends) should I use 1x1 or 2x1 chroma subsampling?
>
> Theoretically it seems like there is enough color information for 1x1,
> but I've never done any side-by-side comparisons.

Hmm, I'd never heard of that... what program gives you that option? Ah,
OK I tried both ways with irfanview and there is a little improvement
turning off color subsampling.
http://www.wfu.edu/~matthews/misc/jpg_vs_gif/JpgCompTest/JpgChromaSub.html
"Chroma subsampling may be a reasonable compression strategy for high
resolution images, but for color web images, one is better served by
turning off chroma subsampling and compressing more aggressively to
achieve the same file size."

'Web' images are usually around 300 pixels wide but your 50% reductions
are probably plenty big to not matter much.

--
Paul Furman Photography
http://edgehill.net
Bay Natives Nursery
http://www.baynatives.com
 >> Stay informed about: Chroma in Downsampled 2x1 JPEG? 

Paul Furman <paul-.RemoveThis@-edgehill.net> asked:
>> When 50% downsampling digital camera images with 2x1 chroma subsampling,
>> (for sending a CD to friends) should I use 1x1 or 2x1 chroma subsampling?
>> Theoretically it seems like there is enough color information for 1x1,
>> but I've never done any side-by-side comparisons.
>
> Hmm, I'd never heard of that... what program gives you that option?

GIMP, ImageMagick, and PSP. Irfanview and Photoshop allow 1x1 or 2x2 only.
In Photoshop you really don't have much control over chroma subsampling;
at quality 7 and above you get 1x1, otherwise 2x2. Likewise in Photoshop
Save-for-Web the dividing line is around 50.

> http://www.wfu.edu/~matthews/misc/jpg_vs_gif/JpgCompTest/JpgChromaSub.html
> "Chroma subsampling may be a reasonable compression strategy for high
> resolution images, but for color web images, one is better served by
> turning off chroma subsampling and compressing more aggressively to
> achieve the same file size."

Thanks for the interesting reference, which studies only 1x1 and 2x1
but neglects 2x1 (used by most digital cameras because they have already
lost chroma information due to RGBG Bayer-pattern sensors.)

> 'Web' images are usually around 300 pixels wide but your 50% reductions
> are probably plenty big to not matter much.

True, chroma subsampling (especially 2x2) looks worst in small images.
 >> Stay informed about: Chroma in Downsampled 2x1 JPEG? 

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Bill Tuthill wrote:
> When 50% downsampling digital camera images with 2x1 chroma subsampling,
> (for sending a CD to friends) should I use 1x1 or 2x1 chroma subsampling?
>
> Theoretically it seems like there is enough color information for 1x1,
> but I've never done any side-by-side comparisons.
>

I don't find much difference aside from when I'm doing graphics work and
have _very_ saturated reds.

- --
Brendan Gillatt
brendan {at} brendangillatt {dot} co {dot} uk
http://www.brendangillatt.co.uk
PGP Key: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0xBACD7433
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (MingW32)

iD8DBQFG+B8FkA9dCbrNdDMRAjf/AKCEViFsUEyShXJO85mpT1bkZwB4mgCfcHCI
at/myrNcTrU+PD3YLZccHAw=
=v9OS
-----END PGP SIGNATURE-----
 >> Stay informed about: Chroma in Downsampled 2x1 JPEG? 

Bill Tuthill <ccreekin DeleteThis @yahoo.com> writes:
>When 50% downsampling digital camera images with 2x1 chroma subsampling,
>(for sending a CD to friends) should I use 1x1 or 2x1 chroma subsampling?

>Theoretically it seems like there is enough color information for 1x1,
>but I've never done any side-by-side comparisons.

How large is a pixel in angular terms as seen by the eye? This depends
on print/display pixels per inch, plus viewing distance. The human
eye's resolution for chroma is about 10 times worse than its luminance
resolution, and there's little point providing more resolution than the
eye can see at the closest normal viewing distance.

For example, if you're making prints at 600 PPI viewed from 10 inches,
you can theoretically resolve up to about 50 cycles/degree, and 35
cycles/degree is practical. With 2:1 downsampling, chroma would be
limited to 25 cycles/degree - but you can't see more than about 6
cycles/degree. So go ahead and downsample; you'll never see the
difference.

On the other hand, if you're displaying on an old monitor at 72 PPI, the
resolution limit is only 6 cycles/degree. Chroma downsampling would
reduce chroma resolution to 3 cycles/degree, and you *can* see that
difference.

Dave
 >> Stay informed about: Chroma in Downsampled 2x1 JPEG? 

Dave Martindale <davem RemoveThis @cs.ubc.ca> wrote:
>
>> When 50% downsampling digital camera images with 2x1 chroma subsampling,
>> (for sending a CD to friends) should I use 1x1 or 2x1 chroma subsampling?
>
> How large is a pixel in angular terms as seen by the eye? This depends
> on print/display pixels per inch, plus viewing distance.

Most people I know use a 1280x1024 LCD display that measures ~ 90 ppi.

> The human eye's resolution for chroma is about 10 times worse than its
> luminance resolution, and there's little point providing more resolution
> than the eye can see at the closest normal viewing distance.

That's the theory behind JPEG, but in practice, chroma subsampling often
produces blocky artifacts in continuous-tone areas around sharp edges
of detailed objects. What surprised me is how bad these blocky artifacts
look in B&W (grayscale) images! You'd think B&W lacks chroma.

> For example, if you're making prints at 600 PPI viewed from 10 inches,
> you can theoretically resolve up to about 50 cycles/degree, and 35
> cycles/degree is practical. With 2:1 downsampling, chroma would be
> limited to 25 cycles/degree - but you can't see more than about 6
> cycles/degree. So go ahead and downsample; you'll never see a difference.

If anyone wants a print, I can go back to the out-of-camera original JPEG.

> On the other hand, if you're displaying on an old monitor at 72 PPI, the
> resolution limit is only 6 cycles/degree. Chroma downsampling would
> reduce chroma resolution to 3 cycles/degree, and you *can* see that
> difference.

So at 90 PPI the resolution limit it 7.5 cycles/degree, and 2x1 chroma
subsampling would reduce (horizontal-only) chroma resolution to 3.75
cycles/degree? I decided to use 1x1 (no chroma subsampling) but someday
I should do some comparisons.
 >> Stay informed about: Chroma in Downsampled 2x1 JPEG?