Join Date: Sep 2005
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
|The yellow/blue crude assessment is obviously not as good as using Photoshop to analyse the jpg file from the camera.|
|A photograph taken of a white sheet of Letter paper under a 6500K fluoro tube would give you the reference photo for the custom WB if your camera supports it (many do).|
|The eyedropper tool then will analyse the white area in the image and give you the relative R,G,B levels that you can calibrate against.|
|but it's *much* better than by eye.|
|I'm cringing in my corner watching this go spinning off the road and into the drainage ditch.
1. The gray card which Kodak produces does not have the same color independent of the light illuminating it. What is special about the card is that the pigments used are such that it usually does not alter the color of the light reflecting off of it. In other words, its color varies with the color of the incident light. You cannot assume it is always glows at a magical D65, 5500K or other color of white. You must also illuminate it with a D65 light source.
2. Outdoor sunlight is not D65. Certainly NOT direct sunlight. D65 is roughly a mix of northern skylight + direct sunlight, but isn't that exactly. It is defined as a specific mix of red, green, and blue stimuli to the eye.
3. Every digital camera includes color filters which are used to separate out the three colors in a scene. Unfortunately, those three color separation filters are not necessarily the same from camera to camera. The differences in filter response are especially troublesome when you deal the the spikey spectral output of digital projection. CRT's aren't as bad but the red phosphor is also spikey in spectral output. This means that the same ratio of RGB won't be seen if you change camera brands.
4. Most cameras white balance the scene. This will also vary the ratios you read of red, green, blue. You may be able to store a manual white balance. That WOULD be useful to teach the camera the ratio which is D65 if you also have an accurate D65 light source to serve as your standard illuminant.
5. 6500K is not D65. D65 is slightly off the Plankian curve but near 6500K. The correlated color temperature indicated on a fluorescent bulb as 6500K doesn't mean it is the right color. It's possible to have a CCT of 6500K and be considerably too green or too magenta. You'll get quite close to the right color if you have a bulb which has a phosphor set designed to match D65. The GretagMacbeth bulb in the IdealLume lamp is probabaly the closest to the right color you will find. Your local pet store's bulbs even though claimed to be 6500K with a high CRI is likely still off the mark. Then again, accuracy may not be that important to you if you are actually pursuing doing grayscale this way.
6. Digital cameras which are "calibrated" at the factory aren't set for shooting under D65 lighting. They are set up to work under daylight, flourescent, etc. so you can't expect RGB to be equal when the photographed image is D65.
It all boils down to needing at least one accurate source of light which is the right color so you can obtain the initial ratios which indicate D65 in your camera. Without that reference, you're defining things in a circular manner.
|the camera idea can work, but you HAVE to get a reference light and shine on a Kodak gray card and VISUALLY make gray on your display match the color you see on the card. Then teach the camera that ratio which is now coming off your display as being the desired color. The spectral selectivity of the color filters in the camera would then be compensated for by the selectivity of your vision (which is what the CIE sensitivities in a color meter mimic)
You can probably demonstrate for yourself the problem of simply letting the camera learn a light as the right color of white. Make your laptop, your projector and D65 illuminated gray card all show grays which are visually identical. Then take a picture which includes all three. I'll bet you that the camera won't see all three as having the same color even though it looks right to your vision. By the inverse token, things which the camera sees at the same color won't necesarrily look the same color to a human.
|You really should set the projector's gray to match the card and lamp VISUALLY first. Then use the reading from the screen and projector to establish the camera's reading for the correct color. Doing it off the card + lamp doesn't correct for the camera's tendency to see slighlty differently from human vision.|
Originally posted by Lifter
With a color corrected monitor, you could then use that to calibrate the digital camera, check it for RGB discrepancies, and adjust as necessary or at least recognize it's problems and compensate. Throw up a full-screen solid 155:66:141 (or whatever) image on your PC monitor, take a picture w/ the camera and check to see how close the numbers match.
Unfortunately, that wouldn't allow the camera to set your projector to the same color. It could only do so if the color filtering in the camera matched the CIE observer curves exactly. They don't or at the very least would be very unlikely to be an exact match. That means colors which look identical to the camera won't necessarily look identical to a human. While doing the above would be perfect for adjusting your projector so your camera maximally enjoys the picture, the human observers will find the grayscale and color a bit off.
Your best, low cost solution to grayscale is to obtain a kodak gray card and a very good reference light. By varying the distance between the card and lamp you can adjust the intensity. Provided you have normal color vision, your eyes are extremely good at picking out differences in two simultaneously viewed colors if they are at the same intensity. Hold the card so it obscures half of a grayscale window and allow the lamp light to strike the card but not the screen. You'll need something to shade the screen from the lamp. Adjust lamp to card distance until the light intensities match. Leave one channel alone (usually green) and adjust the other two channels to make the color of the screen and the card look identical. With practice, you'll find that you can pick up an error of single click of the gain controls. The darker end, visually setting cuts is tougher. Your color vision isn't quite as good at the dark end. Then again, that also suggests (gasp) that the very bottom end grayscale isn't quite as critical since you can't perceive problems there that well any way.
The camera idea seems a good one for looking at tracking. If you can establish the top end accurately using your makeshift "optical comparator," then the camera may allow you to get the dark end more accurately than by eye. It's a thought worthy of experimentation. Just make sure you get at least one good, correct ratio. That you'll have to do by visually color matching one high intensity window to a gray card illuminated with an accurate D65 source.
|We'd still have a problem with the spectral power distributions of my phosphors not exactly matching the ones in your projector. The SPD interacts with the filter set in the camera to produce the energy levels seen by the camera. Unless those filters in the camera are unusually good matches to the CIE observer sensitivity curves reading my display won't solve the problem. It's a problem with two unknowns, both of which vary (the SPD of the display and the camera filter responses). Unless the filter response is a perfect match to the CIE observer curves or you have correction matrices for that filter set for each type and age of display you can't use the camera to establish your standard color. You CAN probably use it to verify tracking after you fix the standard color white visually against a known, correctly colored D65 light source.|
|We actually went down that road...all the way.
We made a prototype software tool that used a tripod-mounted digital camera, which was automated via the USB port to take digital photos of known reference test signals. The photos were downloaded to the computer (automatically), and then were analyzed/averaged pixel-by-pixel to determine how the actual output correlated to the expected output.
When we corrected the software for the camera, the prototype worked reasonably well...for that single camera. It was a tremendous amount of work to get a system that worked with a single digital camera. However, when we tried the same system with 2 other cameras of different brands, it failed horribly...and it wasn't just the automation, but the digital photos themselves varying substantially with regards to color values with the same exact image presented in the same exact environment.
There are CCD-based standard-observer-balanced accurate colorimeters from Radiant Imaging, but those start at $40K and go up from there. Not exactly a typical digital camera. Other than that, a digital camera's CCD filter set would really need to be calibrated to the spectral output of the bulb (or electron gun) of the display device (just like any colorimeter) to work. There isn't a shortcut for this step if accurate colorimetry is the goal.
I would suggest that calibrating with Avia and using just your eyeballs would be preferable to trying to incorporate a digital camera into the mix. Plus, it is much less expensive and will probably get you "close enough" to make you happy.
|Through iteration one can get as close as one likes to the ideal which is like in photo 2. I hope this is clear now.|
|Oh, and there's no Kodak gray card which is mentioned as an integral part of the quotes above.|
Originally Posted by ChrisWiggles
...It is, in my opinion, dangerous for you to suggest this as people will read your statements and think that they can approach D65 using a camera to measure relative values or R, G, and B. ...
My new favorite game is Save The Titanic
Originally Posted by Ericglo
Wouldn't it be better just to use your technique and the description in the "How to eyeball" thread?