Re: CFD: zXIF 2017-0207

John Bowler <[email protected]> Fri, 10 Feb 2017 20:07:47 -0800
Newsgroups gmane.comp.graphics.png.general
Message-ID <CAP7U399+G0deTrFm-Z83o3UkLonW8k94R7CWvH+NoaB5KMc-gA@mail.gmail.com>
On Fri, Feb 10, 2017 at 9:39 AM, Cosmin Truta <[email protected]> wrote:
> Pavel, there is zero amount of overlapping between EXIF and
> gAMA/cHRM/sRGB/iCCP. This is by design.

Cosmin's statement is simply not true.  There is a whole lot of stuff
in CIPA-2016 which is related to the encoding of the image
information.  In the digital camera JPEGs I looked at most of this has
already been handled and the original sensor data has been
perceptually mapped to sRGB, but what the camera recorded was an EXIF
tag that said that the JPEG encoding was sRGB.  I can't remember a
profile.

> As a proof, take a look at the JPEGs or TIFFs coming out your
> smartphone or digital camera. There are very high chances that you
> will see both an EXIF structure and an ICC profile.

Ok; what is the percentage?  Of the images I posted, or, for that
matter, of Glenn's original images, how many have embedded ICC
profiles and how many do not?  You can answer or, anyone can answer (I
posted my samples); I haven't checked.  sRGB came out of the Japanese
manufacturers too, and they knew what they were doing.

But, ultimately, you miss the point; the EXIF data from at least some
cameras contains information that allows the original information to
be reconstructed.  This is contained in the original white point and
the original luminance levels - the absolute light levels.  With that
information it is possible to combine two images from different
sources in a meaningful way semi-automatically, without it the result
just looks weird.

The original information is invalidated when someone starts messing
with the image.  It's invalidated if the best graphics artist on the
planet comes along, looks at the image in the exact sRGB viewing
environment, and changes it so that it is a better sRGB image.

Glenn's approach (the one Greg graciously but incorrectly ascribed to
me) of making the decoder throw out anything that it does not *KNOW*
is right is the fix for this.  The best graphics artist in the world,
when confronted with the information that the original picture was
taken on Vanuatu during a solar eclipse on January 18th would say, "So
what?"  The information needs to be thrown away because it is
hopelessly misleading.

John Bowler

I signed off there because the rest of my response would have been
incredibly irrelevant PNG specific stuff: there are major issues with
cHRM and gAMA and, before v4 with the ICC specification when the data
is used in an image editing environment.   I've given up discussing
this stuff since the last argument over whether sRGB images should
ever be displayed with a blue cast; unlike Willem I didn't stop
contributing to the list I just directed certain peoples' comments
into the bit bucket.

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, SlashDot.org! http://sdm.link/slashdot