Re: Colorimetry in EXIF

This all matches my understanding, with one extra clarification as to why
there is missing information when images from different original scenes are
combined.

At present I'm in agreement that eXIf should be safe to copy; the process I
describe below where the original information is used will still work after
modification of the image just so long as the app that modifies it uses the
relevant PNG chunks to record the new colorspace (sRGB or iCCP).

Having thought about it for a couple of days I'm pretty sure than the
*output* colorspace information is ok as a default but is normally
overridden by explicit or implicit information in the *current* image
format while the original camera scene information is more in the nature of
a comment about the scene than a precise engineering/CIE definition
*unless* you known that the image hasn't been tweaked.

The long explanation:

On Mon, Jan 2, 2017 at 9:20 PM, Cosmin Truta <[email protected]> wrote:

> I am by no means an EXIF expert, but from what I know, the colorimetry
> is about the sensor. Once you have the JPEG, (or TIFF, or PNG) it
> doesn't matter anymore. If you edit it further, that doesn't count,
> either.
>

Yes; the camera takes the sensor data and adjusts it from the measured
brightness (luminance level) and white point (which is assumed to be the
white point the viewer is adapted to) to that of, typically, the sRGB
reference.

In the ideal world of exact engineering this maintains the CIE colors
(CIEXYZ values if you like) in the original scene; a tri-chromat human in
the sRGB reference environment with an sRGB reference monitor will perceive
the same CIEXYZ values (colors) as he would in the original environment.
 (So long as the human matches the CIE spec ;-) - tri-chromat, probably
male, probably northern European.)

The sRGB colors can then be correctly adapted to any other known
environment.

However what is lost in this transform is the ability to merge multiple
images from different original conditions.  For example if I have a picture
of a human in a typical north American living room viewing a picture on a
television with a window showing the outside in the background *I* (I'm an
engineer) do not have the information to modify the picture on the
television or the outside scene to, say, a beach scene and a snow scene
respectively.  I don't know how to change the CIEXYZ values (the colors) I
have for the two new scenes to the actual viewing environment.

This problem arises because the three pictures I have have CIEXYZ values
for a human who is adapted to the light in the three separate scenes and I
don't know what that that light is (luminance and white balance) if I have
just the standard PNG information.  I need to color correct those scenes to
reflect how a viewer adapted to the living room (the main picture) will
perceive a picture of the beach scene on the television (it will be bluish)
and the outside scene (a snow scene, it will probably also be bluish).

Well, OK; this is about doing such merging in computer software with no
user involvement.  In practice a graphics artist just tweaks the colors
until things look right.  Since the output is a set of colors and colors
are entirely defined by the human brain this tends to work very well.

The only time anyone cares is when merging original digital data into a
common color space; then it is necessary to convert from an sRGB
(typically) encoding to a common one that correctly represents colors
outside the sRGB gamut and luminances outside the sRGB range.

This is only likely to happen when using an image processing pipeline that
works via a universal extended color space and I would expect this to
happen as the first step of the process.  At that point the original color
information is factored out and an app that used it would probably not
maintain it in the universal format image data.

So, in reality, the original scene information recorded by the camera is
more like a comment ('this was a beach scene') than something with
engineering relevance *unless* you know you are dealing with the original
data.

John Bowler <[email protected]>

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

_______________________________________________
png-mng-misc mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/png-mng-misc