[IPOL discuss] ICC color profiles

[Pascal Getreuer]

Dear all,

Miguel is correct that PNG images may contain an ICC color profile or
gamma correction exponent that modifies the way that the image looks.

Unfortunately, the PNG standard included the regrettable advice to
*guess* the gamma exponent value of the user's machine, which has led
to confusion and inconsistency in implementation of these features.
Some programs write PNG files with blindly-guessed color
profiles/gamma exponents, so the PNG looks wrong when displayed
according to this bad information.  It is controversial whether PNG
color profiles and gamma should be respected at all [1] [2] [3].  From
my own experience, GIMP and MATLAB do not recognize gamma in PNG, but
Firefox does.

Supposing we can trust the color profile/gamma, it would be desirable
for IPOL to transform the input image intensities to a single
well-defined color space, such as sRGB [1] (the sRGB color space is
the most widely accepted model of how intensity values correspond to
physical colors when rendered by a typical computer display).  In this
way, the subsequent processing is agnostic to the source color space.
Output images should always be written as sRGB.

But what if the color profile/gamma is wrong?  How can we decide when
to trust it?

If anyone can shed some light on this topic, I would love to hear it.

Thanks,
Pascal

[1] http://hsivonen.iki.fi/png-gamma/
[2] http://morris-photographics.com/photoshop/articles/png-gamma.html
[3] http://www.firewing1.com/node/14
[4] http://en.wikipedia.org/wiki/SRGB


On Mon, Dec 10, 2012 at 6:38 AM, Miguel Colom <colom at cmla.ens-cachan.fr> wrote:
> Dear all,
> these days we've been dealing with a weird problem with the Noise Clinic
> demo (still in development, not published). We denoised an image and the
> result was too dark. However, the values of the pixels very quite similar.
>
> We found that this was caused since the input image had a ICC color profile
> [1] "ACER X233H" and the denoised image doesn't have any. Therefore, they
> show different.
>
> Perhaps it'd be a good idea in the future to modify the io_png helper files
> to treat somehow the ICC color profiles. Or even better, that the demo
> system does.
>
> Best,
> Miguel
>
> [1] http://www.libpng.org/pub/png/book/chapter10.html#png.ch10.div.7
>
>
>
> _______________________________________________
> discuss mailing list
> discuss at list.ipol.im
> https://tools.ipol.im/mailman/listinfo/discuss