JPEG and size of colorspace

But surely this is exactly what they fixed -- although not with the interface he suggested -- in SPP 1.1.

From Sigma web site:

"Foveon and Sigma appreciate the attention that users give to the details of the SPP software. Several users noticed that the v1.0 Sigma Photo Pro saved JPEG files with 2x2 subsampling of chroma channels, which degraded color detail in images relative to files saved as TIFF. In SPP update v1.1, JPEG files include full-resolution chroma channels whenever the JPEG quality is set to 7 or higher."
 
I must say I really don't think that JPEG should give
stair-stepping, ever. If the chroma is sub-sampled properly and
then interpolated properly where should be no stepping. Just a
fairly blurry chroma channel, which isn't too much of a problem.

Anyone know the reasons?
Unless the decompression algorithm happens to know precisely how the channel was subsampled, interpolation could blur detail that the algorithm otherwise had a chance of reconstructing exactly.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/
 
Erik, I'm not familiar with that non-colorimetric usage of "color
space". Perhaps you can provide a typical reference.
It's easy. Just google for "colorimetric color space" (include the
quotes). If there were not a distinction, then this combination
should almost never appear because it would be redundant. Also
look for "device dependent color space" because these are almost
never perfectly colorimetric.
So, I did as you suggest, and google only found a handful of hits. I agree that google is a pretty good tool for suggesting a position, but let's not pretend it can settle an argument this way, or else I win by default. You said the literature is full of color space used in a non-colorimetric sense, so provide an example, OK?
Perhaps I'm being too strict in trying to conceptualize sensor
space as something different from a color space
Try here:
http://www.dpreview.com/reviews/minoltadimage7/page13.asp
That just shows that Minolta put out images in an unspecified space.

j
 
So, I did as you suggest, and google only found a handful of hits.
You're correct, I must have not noted my search term very well.
You said the literature is full of color
space used in a non-colorimetric sense, so provide an example, OK?
Do a literature search: http://citeseer.nj.nec.com/ and try colorimetric and color space. You can also check for citations to Luther and Ives work and see how people use it. My real nitpick with your posting is that Luther and Ives work defines "colorimetric" and not "color space" (which does not seem to have a definitive cite definition.)

Really, this is nitpicking over whether "color" is purely a property of human vision (and hence the only "color spaces" are the CIE standard models and anything colorimetric with them) or if it's more general in that any visual system has "color." (Do bees have a "color space" or do you call it a "bee vision space"?) If you, like Joe, want to be strict and say that "color" is by definition only human and use other terms for other spaces, fine. However, it's useful to have a term that applies to all possible "visual frequency spaces" even those that are not colorimetric (i.e. perfect linear transformation to human color space.)

--
Erik
 
Within the volume of the colorspace, the encoding, whether TIFF or
JPEG, 8 or 16 bit, determines the accuracy with which individual
pixels (for TIFF) or blocks of pixels (for JPEG) are represented.
When the accuracy is reduced, by encoding to fewer bits, then fewer
unique colors will be found in an image, as sg10 keeps telling us.
JPEG compresses 8-bit lossless files at about a 3 to 1 ratio. JPEG compresses 16-bit lossless files at about 6 to 1, the increase in compression ratio is attributable to the colorspace reduction from 48-bit to 24-bit color, which is the inherent limit of the JPEG format. That is somewhat deceiving because a 16-bit TIF can accommodate 48-bits color but only 36-bits are used if camera RAW is 36-bit.
But can one see the difference? Sometimes yes, sometimes no. I
don't think anyone can show us a difference between an 8b TIFF and
a 16b TIFF, since the 16b has to be converted to 8b to be put a on
a screen or sent to a printer, in most cases. When JPEG cuts the
file size by a factor of 10 and the number of unique colors by a
factor of 4 or more, can you see the difference relative to an
original TIFF? Not usually, though sometimes with careful
inspection some differences may be findable.
You can't see that difference on a 24-bit color monitor, but you can print the difference on a 36-bit color printer.
 
JPEG compresses 8-bit lossless files at about a 3 to 1 ratio. JPEG
compresses 16-bit lossless files at about 6 to 1, the increase in
compression ratio is attributable to the colorspace reduction from
48-bit to 24-bit color, which is the inherent limit of the JPEG
format. That is somewhat deceiving because a 16-bit TIF can
accommodate 48-bits color but only 36-bits are used if camera RAW
is 36-bit.
jpeg supports up to 12 bpp, but it's rarely implemented.

The conversion between the bit depth of your RAW file and the number of bits used in a TIFF will depend upon the physical properties your sensor and the gamma of the target colorspace.
You can't see that difference on a 24-bit color monitor, but you
can print the difference on a 36-bit color printer.
I think you will have a hard time finding a color printer and a set of viewing conditions that will produce visible distinctions. If you believe Poynton's well regarded gamma FAQ, with an appropriate gamma 9 bits is more than enough to cover the full range of intensities we can perceive. The range you could cover with a printer would less since the printer will not be able to produce the blackest black or the whitest white we can perceive.

--
Ron Parr
FAQ: http://www.cs.duke.edu/~parr/photography/faq.html
Gallery: http://www.pbase.com/parr/
 
You can't see that difference on a 24-bit color monitor, but you
can print the difference on a 36-bit color printer.
Exactly what is a 36-bit color printer? Can it actually print 68 billion different shades? And even if it, could how could you tell?

--
Erik
 
JPEG compresses 8-bit lossless files at about a 3 to 1 ratio.
"8-bit lossless file" in an oxymoron.

And JPEG can give you any compression ratio you want, subject to a quality tradeoff.

Maybe you meant that JPEG has a lossless mode that can compress 8-bit files by 3 to 1? Not sure. The "normal" JPEG and "lossless" JPEG are completely different algorithms, and the lossless one is seldom used. The "normal" one at top quality is not quite lossless, and typically around 3 to 1 or better, so maybe that's what you are referring to?

j
 
JPEG compresses 8-bit lossless files at about a 3 to 1 ratio.
"8-bit lossless file" in an oxymoron.
Yes, you are right in the context of 36-bit color. Although it is possible to losslessly store or compress an 24-bits of starting information with 8-bit file formats.

Your 'oxymoron' is why the JPEG compression ratio goes up when you start with 36-bit color, the color has to be converted from 36-bit to 24-bit before JPEG starts. Both steps are lossy and both steps stem from a decision to JPEG. A lot of people don't think about that.
And JPEG can give you any compression ratio you want, subject to a
quality tradeoff.
Maybe you meant that JPEG has a lossless mode that can compress
8-bit files by 3 to 1? Not sure. The "normal" JPEG and "lossless"
JPEG are completely different algorithms, and the lossless one is
seldom used. The "normal" one at top quality is not quite
lossless, and typically around 3 to 1 or better, so maybe that's
what you are referring to?
That is right too, I was talking about using the highest quality setting of normal JPEG, since that is the one used here.
 
You can't see that difference on a 24-bit color monitor, but you
can print the difference on a 36-bit color printer.
Exactly what is a 36-bit color printer? Can it actually print 68
billion different shades?
Yes they can produce all 68B shades at about 4000 dpi. The 36-bit Lightjet and Lambda model costs about about $130,000 each, but using themthrough a service isn't that bad, maybe $5 per sqaure foot in small sizes. Photo printer really doesn't mean photo-quality at that level, these produce actual photographs using light exposed silver halide paper. Try http://www.ebluce.com/
And even if it, could how could you tell?
Fot that much money, I hope you can tell!
 
You can't see that difference on a 24-bit color monitor, but you
can print the difference on a 36-bit color printer.
Exactly what is a 36-bit color printer? Can it actually print 68
billion different shades?
Yes they can produce all 68B shades at about 4000 dpi. The 36-bit
Lightjet and Lambda model costs about about $130,000 each,
I seriously doubt this can actually be achieved on any real media. First, from the specifications page for the 5000XL
Color Resolution: 36 bits; each 8-bit input channel is
interpolated to 12 bits to produce a 36-bit output color space,
resulting in smoother gradients, and increased highlight and
shadow detail
It's not even actually 4000dpi. Instead:
Continuous-tone: 200, 300 dpi, apparent half-tone resolution is 4,000 dpi
I'd prefer to see some independent analysis of the output rather than marketing material. I suppose you think that most scanners actually have a Dmax of their full bit depth as well?
Fot that much money, I hope you can tell!
It may be the best printer out there, but then again I doubt that any "24-bit" printer can actually get all 16 million colors either.

--
Erik
 
You can't see that difference on a 24-bit color monitor, but you
can print the difference on a 36-bit color printer.
Exactly what is a 36-bit color printer? Can it actually print 68
billion different shades?
Yes they can produce all 68B shades at about 4000 dpi. The 36-bit
Lightjet and Lambda model costs about about $130,000 each,
I seriously doubt this can actually be achieved on any real media.
First, from the specifications page for the 5000XL
Color Resolution: 36 bits; each 8-bit input channel is
interpolated to 12 bits to produce a 36-bit output color space,
resulting in smoother gradients, and increased highlight and
shadow detail
It's not even actually 4000dpi. Instead:
Continuous-tone: 200, 300 dpi, apparent half-tone resolution is 4,000 > > dpi
I'd prefer to see some independent analysis of the output rather
than marketing material. I suppose you think that most scanners
actually have a Dmax of their full bit depth as well?
The reason the dpi is listed as 'apparent' is because dots are not mixed to produces tones. Instead red, green, and blue laser light is mixed to produce colored light to match the bit description. Silver halide is exposed then developed in a traditional sense. There are a several machines that can develop true photographic emulsions from digital files. You might be right and photographs can't really show all conceivable colors, I don't know.

Scanning is different.
Fot that much money, I hope you can tell!
It may be the best printer out there, but then again I doubt that
any "24-bit" printer can actually get all 16 million colors either.
I think you are probably right about mixing inks on a surface, but I'm not sure about that either.
 
So, I did as you suggest, and google only found a handful of hits.
You're correct, I must have not noted my search term very well.
You said the literature is full of color
space used in a non-colorimetric sense, so provide an example, OK?
Do a literature search: http://citeseer.nj.nec.com/ and try
colorimetric and color space. You can also check for citations to
Luther and Ives work and see how people use it. My real nitpick
with your posting is that Luther and Ives work defines
"colorimetric" and not "color space" (which does not seem to have a
definitive cite definition.)

Really, this is nitpicking over whether "color" is purely a
property of human vision (and hence the only "color spaces" are the
CIE standard models and anything colorimetric with them) or if it's
more general in that any visual system has "color." (Do bees have
a "color space" or do you call it a "bee vision space"?) If you,
like Joe, want to be strict and say that "color" is by definition
only human and use other terms for other spaces, fine. However,
it's useful to have a term that applies to all possible "visual
frequency spaces" even those that are not colorimetric (i.e.
perfect linear transformation to human color space.)
Really, I was just nitpicking you over whether you could provide support for your assertion that my strict definition of colorspace is not what is commonly used in the literature. If you can't, that's OK, but don't put the burden on me to go do the research to prove myself wrong.

My argument had nothing to do with human vision per se, but with standardized colorimetry (defined relationship to XYZ, if you like), which is something that I claim all colorspaces have (in my strict definition) but which camera spaces do not have. It's a subtle distinction, but was relevant to some point way up-thread...

Even if my definition of colorspace turns out to stricter than what is in use, my point remains that camera space does not have a defined relationship to a colorimetric space such as XYZ. The relationship needs to be chosen according to someone's preferences, and different users will prefer to push the "errors" around to different places.

Actually, that's likely to be true even if a camera's sensor met the Luther-Ives condition. There would then be a defined relationship to XYZ of the subject scene, but that is very likely not what a photographer will prefer in his reproduction.

j
 
If you can't,
that's OK, but don't put the burden on me to go do the research to
prove myself wrong.
There is not a definitive answer because color space does not appear to be a rigorously defined term. One cite would not prove anything either way. I pointed you to the same sources I used in making my decision, e.g., how technical papers tend to use colorspace. In general, the ones I looked at tended to use it with specific modifiers that suggest to me that it's a more general term.
My argument had nothing to do with human vision per se, but with
standardized colorimetry
You cannot separate the two (at least not in the context of Luther - Ives), because standardized colorimetry is all about human vision and human visual distinctions (otherwise it's just spectroscopy.)
(defined relationship to XYZ, if you
like), which is something that I claim all colorspaces have (in my
strict definition) but which camera spaces do not have. It's a
subtle distinction, but was relevant to some point way up-thread...
And remember that my subtle distinction (aka nitpick) reflected in the title of this thread was that Luther-Ives defined "colorimetric" not colorspace. At least, I could find no reference strictly linking the two terms.

--
Erik
 
If you can't,
that's OK, but don't put the burden on me to go do the research to
prove myself wrong.
There is not a definitive answer because color space does not
appear to be a rigorously defined term. One cite would not prove
anything either way. I pointed you to the same sources I used in
making my decision, e.g., how technical papers tend to use
colorspace. In general, the ones I looked at tended to use it with
specific modifiers that suggest to me that it's a more general term.
What sources did you point me at? You haven't yet named a paper where "color space" is used for a non-colorimetric space. I'm perfectly open to believing that they exist, but remain puzzled as to why you don't show me one.

I feel this discussion has gone full circle at least once, and I forgot how we got into it.

j
 

Keyboard shortcuts

Back
Top