The app actually does have all Sherwin Colors, they are just not all in the dropdown (you can search for them). I added all of them in the dropdown for now in a really janky way. I'll do a proper design in the next few days.
I'm curious... if the RGB values are all accurate, is there a standard color and exposure calibration for paint swatches that is shared across brands?
The white point (color temperature) is easy, using whatever light source they're being illuminated with.
But what determines whether a particular white is (240, 240, 240) or (250, 250, 250) or (255, 255, 255)? The exposure seems far more arbitrary. Is there some kind of maximally bright matte "reference white" material photographed next to each swatch, that then gets calibrated to (255, 255, 255) or similar?
If so, it would be pretty cool if you could buy a reference white chip like that, hold it up to a painted wall in your house, and have an app that could take a photo with both and output the "true" current paint color.
The frontend you see is RGB due to the internet and the display you're using. But on the backend (the manufacturer's system, no idea what the web site employs) the color science is all in spectral data. The xyz color matching functions [1] can be used to determine human response to that spectra, and then to 'CIE L* a* b*' for unambiguously referring to the color. The typical white in the paint industry is CIE illuminant D65.
You're right to wonder about "what is white" which gets into topics of 'chromatic adaptation', 'color constancy', and 'discounting the illuminant'.
If you get a perfectly diffuse white tile, what color is it when illuminant by incandescent versus fluorescent? And there are more than a few CIE F illuminants, and more proprietary ones. And hence each application tends to standardize on, and thus assume, an illuminant. You can use some other illuminant if you want, but then you're in custom territory and it wouldn't surprise me if some companies are doing that, secret sauce.
In special cases like automotive and textiles, it's common to see multiple spectral values due to different colors at different angles, and thus spectrogoniometry.
However color matching is more complex than that. "Color of the light source" varies considerably, not only due to the illuminant itself, but also the materials the illuminant is reflecting off of (e.g. walls). And color temperature only defines black bodies, which most light sources are not: you need information about the full spectrum of the illuminant to determine how it interacts with the material to be color-matched. This is getting a bit into metamerism: https://en.wikipedia.org/wiki/Metamerism_(color) (Of course, paint cannot match an arbitrary spectrum. Matching under specific lighting conditions is the least you can hope for.)
And the above is considering color of a surface only as a scalar value, but it's really a function of illumination angle as well, due to gloss. Direct lighting on a glossy surface reflects more of the color of the gloss, whereas oblique lighting reflects than the paint itself. Diffuse lighting is some combination thereof.
(I am not a color scientist by any means, just interested in the topic. Experts please correct!)
Fascinating. I never knew there was an "standard" mapping between real-world colors and RGB.
Now if only online clothing store photos would map to that, so that the olive green coat I order doesn't turn out to be closer to medium brown or vice-versa... :P
I got one of those Color Muse color matching devices and this matching gets hard when you have glossy or rough surfaces. I've been trying to get some touchup paint to cover a stucco patch and despite many attempts, never can match it close. The colors look different at various angles.
I have the same problem. My current plan is to illuminate a patch from many angles to eliminate all the shadows. Add a reference white or gray card. Take a photo then in GIMP, adjust the white point, blur the stucco, read off
the RGB and look up the closest matching color code. I have yet to find a way to look up by RGB.
If it somehow works I’ll avoid having to paint the whole house.
Your best bet is to cut a piece of paint off the wall and take it to a paint store. They probably won’t be able to match it exactly, but it’ll be closer than you can do in RGB.
The colorimeter at your local paint store, looking at a paint chip from your wall, is very likely going to do a whole lot better than the process you just described. If you call the store, ask how big of a paint chip sample they need to match existing, weathered paint. It very well may be tiny (like, 1cm^2), and they should do this color match for free. You could buy just a pint, first, to see that it works.
I've been quite surprised by how non-uniform most digital camera imagers and in-body post processing color pipelines muck with the spectrum coming off the sensor, all in the name of producing a more pleasing JPEG. If you still want to do what you described, I'd use RAW and use dcraw to convert to tiff, and open that in gimp.
The paint you get in a tin varies a little depending on the batch - even if you get exactly the same colour from exactly the same brand - so unfortunately you’ll never be able to get a precise match (pro-tip: if you suspect that you’ll run out of paint whilst doing something, go buy another tin and mix it with the one you already have - it’ll be much harder to tell you ran out).
The device you are using is presumably (from the name) a colorimeter. In the industry we use spectrophotometers to analyse the full spectrum of reflected monochromatic light, allowing the reconstruction of the whole visible spectrum’s composition, rather than merely matching the overall average wavelength. For some surfaces (very glossy, metallic finish, & cetera) multi-axial spectrophotometers are advisable.
Source: family firm is in the paint industry, worked in it for close to fifteen years.
Probably a (mono-axial) spectrometer: the muti-axial devices are usually used for glossy varnishes or metallic finishes featuring either aluminium or mica inclusions. Multi-axial spectrometers are typically an order of magnitude more expensive and are usually limited to industrial (typically automotive) applications.
I've used the following to match RGB and HTML color codes to paint colors: http://www.easyrgb.com/en/match.php. Wasn't perfect, but got me pretty close!
Something is off. The matched colors show an RGB with each, but it's the same value when the color is obviously different.
I took a screenshot of it and pulled it up in an image editor to make sure.
How do you calculate the distance between these colors? RGB distance will be close enough, probably, but for matching paint/ink in the real world, it's important to consider visual (sensation) difference.
34 comments
[ 3.5 ms ] story [ 83.5 ms ] threadAny intentions on adding suggested palettes?
That way the app could adjust for fading and tell me what color that paint was originally.
The white point (color temperature) is easy, using whatever light source they're being illuminated with.
But what determines whether a particular white is (240, 240, 240) or (250, 250, 250) or (255, 255, 255)? The exposure seems far more arbitrary. Is there some kind of maximally bright matte "reference white" material photographed next to each swatch, that then gets calibrated to (255, 255, 255) or similar?
If so, it would be pretty cool if you could buy a reference white chip like that, hold it up to a painted wall in your house, and have an app that could take a photo with both and output the "true" current paint color.
You're right to wonder about "what is white" which gets into topics of 'chromatic adaptation', 'color constancy', and 'discounting the illuminant'.
If you get a perfectly diffuse white tile, what color is it when illuminant by incandescent versus fluorescent? And there are more than a few CIE F illuminants, and more proprietary ones. And hence each application tends to standardize on, and thus assume, an illuminant. You can use some other illuminant if you want, but then you're in custom territory and it wouldn't surprise me if some companies are doing that, secret sauce.
In special cases like automotive and textiles, it's common to see multiple spectral values due to different colors at different angles, and thus spectrogoniometry.
[1] https://en.wikipedia.org/wiki/CIE_1931_color_space
However color matching is more complex than that. "Color of the light source" varies considerably, not only due to the illuminant itself, but also the materials the illuminant is reflecting off of (e.g. walls). And color temperature only defines black bodies, which most light sources are not: you need information about the full spectrum of the illuminant to determine how it interacts with the material to be color-matched. This is getting a bit into metamerism: https://en.wikipedia.org/wiki/Metamerism_(color) (Of course, paint cannot match an arbitrary spectrum. Matching under specific lighting conditions is the least you can hope for.)
And the above is considering color of a surface only as a scalar value, but it's really a function of illumination angle as well, due to gloss. Direct lighting on a glossy surface reflects more of the color of the gloss, whereas oblique lighting reflects than the paint itself. Diffuse lighting is some combination thereof.
(I am not a color scientist by any means, just interested in the topic. Experts please correct!)
https://en.wikipedia.org/wiki/Middle_gray
Fascinating. I never knew there was an "standard" mapping between real-world colors and RGB.
Now if only online clothing store photos would map to that, so that the olive green coat I order doesn't turn out to be closer to medium brown or vice-versa... :P
If it somehow works I’ll avoid having to paint the whole house.
Take your samples to a good paint store. Those machines are more sophisticated then you may think.
I've been quite surprised by how non-uniform most digital camera imagers and in-body post processing color pipelines muck with the spectrum coming off the sensor, all in the name of producing a more pleasing JPEG. If you still want to do what you described, I'd use RAW and use dcraw to convert to tiff, and open that in gimp.
Source: family firm is in the paint industry, worked in it for close to fifteen years.
Edit: actually it’s almost possible to do what I need. It’s just be much easier if it was possible to find closest match. Still this could be helpful.
e.g. https://www.matchmypaintcolor.com/ppg/medieval-forest shows RGB 1,116,120 for all 3 swatches, but only the Maxi Teal is that RGB, the other tiles have other RGB values.
https://en.wikipedia.org/wiki/Color_difference
What are these? The website doesn't seem to explain anywhere.
Any idea how this process compares?
Compare the pair:
- https://www.matchmypaintcolor.com/ (this site)
- http://surfweb.com/ (fake site)