[argyllcms] Re: help with camera profile

  • From: Ben Goren <ben@xxxxxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sat, 28 Jul 2012 18:31:43 -0700

On 2012-07-28, at 6:08 PM, Stephen T wrote:

> Visual evaluations are not satisfactory because there are dozens of variables 
> that affect whether or not the photo looks "good". You can review many 
> photos, some will be good and some will look wrong. I've wasted hours doing 
> this.

There actually is a visual comparison that can prove quite useful: a blink 

Create a synthetic version of a chart such that the colors match what you get 
when you measure the chart with a spectrophotometer. You can do this in 
Photoshop easily enough, with the caveat that Photoshop only allows integral 
specifications of colors and it doesn't necessarily have the best code for 
translating said values into your working RGB space.

Then, photograph the (real, not on-screen) version of the chart, and run said 
photo through your profiling process.

Put the photo on a separate layer in Photoshop and scale / distort it until it 
overlaps reasonably well with the synthetic version. The difference blend mode 
is helpful for this sort of thing.

Then, set the blend mode back to normal and rapidly toggle the visibility of 
the photo. Differences will stand out quite apparently.

Another caveat: this assumes that the chart colors lie within the gamut of your 
monitor -- or that you don't care too much about out-of-gamut colors.

You can also leave the photo in difference blend mode but don't do the blink 
thing. The darker the patch, the closer the match. That will work regardless of 
the respective gamuts.

Now, combine that with profcheck, and see which patches profcheck has farthest 
from matching and how they actually look. Is the worst offender a near-black 
super-saturated violet that only occurs in shadows whose color fidelity you 
really don't care about? No worries; just ignore it. Is profcheck reporting 
worst-cast differences of 2 DE, but that's in the skin tones, and you're 
wondering if it's a problem? Do the blink comparator check and see what 2 DE 
actually looks like in this case, and decide just how much of a problem it 
really is.

With a really good profile, you'll have to do a lot of flashing back and forth 
before you spot any differences, and you'll have to search for those 
differences again if you look away. With a decent profile, it'll generally look 
the same but there'll be an overall (but subtle) shift in appearance. With a 
bad profile (or no profile) the differences are instantly obvious, and no two 
patches will match.

Oh -- also note that this works even if you don't have the world's best monitor 
profile, since you're comparing the idealized digital version to the 
photographed digital version, not your on-screen version to the actual physical 



Other related posts: