Predictive color temperature model

mattstevenfisher

TPF Noob!
Joined
Jul 17, 2013
Messages
8
Reaction score
0
Location
Ithaca, NY
Can others edit my Photos
Photos OK to edit
I’m a student at Cornell University working on a design project aimed at creating a formulaic model to predict the color temperature of sunlight at any time and geographic location in the world. Our algorithms are well on their way and we need empirical data to verify our work. Thus, we are turning to the online photography communities to help us out!

We are asking people from all around the world to capture the color temperature of direct sunlight on a, preferably, sunny day by photographing (in RAW) a white balance card pointed toward the sun. From that image we’ll determine the color temperature and add it to our growing dataset. While we want pictures taken at any time of day and sun position, the white balance card musts be placed in direct sunlight.

Please send all photos (in any RAW format) to [email protected] and include the following information in the email:
-Location (lat/long, if possible)
-Time
-Weather description (brief: cloudy, clear, overcast, a picture would work as well)

$2013-07-18 13.14.49.jpg
 
Interesting. Is this for commercial application (i.e. improved AWB on GPS equipped cameras)? Or for a class/dissertation, etc.?
 
For neither actually. As an engineer and someone passionate about photography, I thought this was a cool side-project to pursue. I'm not sure exactly what I'll do with it yet, but I'll keep people posted!
 
Why? Every camera I know that has the ability to set the color temp likewise has the capacity to measure it with the click of a button.
 
Ysarex, that is very true. I'm approaching this problem from a more "fundamental research" point of view, rather than photography-in-practice one. I'm curious about creating a predictive model explores how color temperature varies across the world based on lat/long, altitude, weather, etc., and to the best of my knowledge, something like this doesn't already exist.
 
Time of day is only one variable.

Pollution index, humidity, pollen count, season and others I'm not remembering right now also play a large roll in color temp.
 
Seasons are accounted for in the sun models that I'm using (from NOAA) and a lot of other weather and atmospheric conditions, including humidity, are well documented in NOAA databases. From my understanding of various sky models, Raleigh scattering (by small particles) is the predominate factor in determining color temperature and Mie scattering (large particles such as pollen, etc) plays a lesser one. I'm hoping to accumulate a large enough data set that these environmental factors will be statistically negligible when I fit curves to different portions of the overall data set.
 
An interesting project. How will you account for inaccuracies in the values - from camera errors, target variation and non-neutrality, and operator error? Is there a chance that the errors may swamp the true variations? I find different CCT values from the camera, an incident colourimeter (Gossen Color-Pro 3F, Minolta Colormeter II and Colormeter IIIF - none of which agree within their reading precision, even after calibration) and a spectrophotometer. The reading from a colour meter or a spectrophotometer may not be the kelvin value that gives an accurate white balance in camera or in raw conversion, both of which may be different, depending on the raw converter. This is not an exact science.

I assume that your search for previous work has included a check of Wyszecki and Stiles, and the reference section therein. I'm surprised that you didn't turn anything up, since it has been of interest to the CIE since the 1930's.
 
Why? Every camera I know that has the ability to set the color temp likewise has the capacity to measure it with the click of a button.
Yes, but you don't really WANT your camera to do this, if the color cast is part of the actual world and not a faulty initial setting in your camera.

Your camera is making a best guess, and especially when conditions are odd, it can be wrong by quite a bit. If it can communicate with internal software or an external server (in a not-so-distance future of cheaper public wireless), then it can know for sure that the sky is SUPPOSED to be reddish yellow right now due to pollen count and time of day and latitude, and probably won't do as much correcting for that as it might have guessed with primitive blind software.

Whereas color casts that DON'T match the expected data would be more likely to be adjusted. (Of course, you'd have to have a simple to change setting for whether you're outside or not)
 
Why? Every camera I know that has the ability to set the color temp likewise has the capacity to measure it with the click of a button.
Yes, but you don't really WANT your camera to do this, if the color cast is part of the actual world and not a faulty initial setting in your camera.

Your camera is making a best guess, and especially when conditions are odd, it can be wrong by quite a bit. If it can communicate with internal software or an external server (in a not-so-distance future of cheaper public wireless), then it can know for sure that the sky is SUPPOSED to be reddish yellow right now due to pollen count and time of day and latitude, and probably won't do as much correcting for that as it might have guessed with primitive blind software.

Whereas color casts that DON'T match the expected data would be more likely to be adjusted. (Of course, you'd have to have a simple to change setting for whether you're outside or not)

I wasn't referring to a camera's auto white balance function. I was referring to a camera's ability to set a custom white balance from a know target.

Joe
 
Ah well yes. The advantage of the algorithm and system would be that the auto white balance would do much of what you would normally have to take extra steps to do it custom. If successful, of course.
 
Most (if not all) cameras do have that feature. My original plan was to use Lightroom to determine, for myself, the color temperature of the card. That way, I can see the image and the inherent raw data to verify it was taken in a manner than can be included in the dataset. Also, I have found that there are slight variations of white balance over the surface of the card and I wanted to use Lightroom to "average" the reported values.

It would be awesome to have someday a "smart" connected firmware that can use some kind of predictive models as this (but undoubtedly a much more advanced) to predict the expected color temperature and adjust accordingly!
 
So I spent a little while looking into the variation between different lens and body configurations and have found that there is minimal differences in reported color temperature. My project is indeed not an exact science and I'm more curious about over all, large scale trends - in which case a different of a hundred or two Kelvin is negligible and averages out.

I borrowed every piece of camera equipment that I could get my hands on and took some photographs in indoor, controlled lighting in order to remove as many variables as possible. Here are the results:
$WBdiff.jpg

I also looked into different white balance cards, and the differences there are also negligible:
$WCcompare.jpg

That being said, I'm confident in the crowd sourced data and am still looking for submissions if you would like to help out!
 
So you're expecting to see some sort of global-scale variations in color temperature that exceed a 100K?

I am curious as to what the possible causes of such differences might be.
 
First of all, I'm trying to collect as many data points as possible to build a statistical model that can verify the analytical model that we have pieced together from a variety of sky models (Bird, Carter, Riodan, etc.) - to do that effectively, we need a large data set. Also, all of the data points that I collected myself were from Ithaca, NY (upstate NY) which has little to no pollution and is at 800 ft elevation. I'm curious to see how elevation effects color temperature (someone in Denver/Quito/Santiago would be perfect for this. I'm predicting that light is cooler in higher altitudes as there is less atmosphere scattering blue light) as well as testing the model in winter (i.e. southern hemisphere now). I'm going to do a series of analysis once I get a full dataset and then see what kind of trends I can extract. Hope that all makes sense
 

Most reactions

New Topics

Back
Top