"Daylight" fluorescents for auto white balance

[Misty]

I've been looking at "daylight" fluorescent lights for use with a book scanner. I know most fluorescents produce pretty bad results, but I've been impressed with what I've found from the daylight bulbs so far. The bulb I'm using is the 26W (100W equivalent) GE Daylight bulb - their page for it is here.

They're designed to produce a white, natural light, and in my experiments Canon's AWB has produced very nice results even though it's usually very bad with fluorescents. Even better, I've also found that they produce a standard temperature. They're rated at 6500K, and while I haven't formally tested the temperature I've found that manually setting the white balance temperature to 6500K is producing perfect or close to perfectly accurate colour. That works best when shooting in raw, naturally.

Here's a few examples:

Poster photographed, colours from camera's builtin AWB



Same photo, set manually to 6500K - this is accurate to the actual poster. You can see that there's only a very minor difference.



Despite the fact that the photos were taken in very non-optimal lighting environments (my dining room, obviously not designed for book scanning), the colour results are very good.

I have another couple of examples, but they're taken from a laserdisc cover that is within copyright. Is that all right to post?

[daniel_reetz]

I have another couple of examples, but they're taken from a laserdisc cover that is within copyright. Is that all right to post?

Rob and I have talked about such things, we do try to keep the copyrighted material to a minimum. But if you're using it to illustrate a point, and comment on it, it should be within Fair Use standards. I say, go ahead if it will better illustrate the experimenting you've done (which, BTW, looks great, thanks for sharing).

[Misty]

All right, thanks Daniel.

It goes without saying that this is a silly image, but it's a pretty good example as far as colour reproduction goes and it has very different tones than the poster did. Here's an example as shot by the camera under the daylight CFLs in the same conditions as that poster, with the white balance temperature set to 6500K.



For comparison, here's a reference image shot in different conditions which has been heavily manually colour corrected for exact accuracy. You can see that there's actually very little difference between the two images, and most difference has to do with the shooting environment and clipping for the pure whites/blacks.



That said, Canon's AWB did a bad job on this image - it's similar, but has a noticeable blueish cast. I think it's because there's very little recognizable white on the cover and the background is inconsistent, whereas the poster had enough white areas for the sensor to locate. However, the WB presets for Cloudy and Daylight both provide quite nice results. Cloudy provides overall the most accurate results, though you can see that there is a minor colour cast that is concentrated mainly in yellows.



Given the consistent colour temperature there's probably no reason to have to rely on the in-camera AWB, unless you're using a camera that doesn't shoot raw or support CHDK. Even then, the WB presets will give very acceptable results.

For my cradle, I'm going to be using cameras that shoot raw natively - Canon G10s, which I snapped these photos with. I'm not too familiar with CHDK's functionality, so I'm curious as to whether people going that route can program the temperature in-camera or if that needs to be done using CHDK-generated raws. Either way it should be able to be automated using free software, given that the ideal values are exactly the same for every image.

[spamsickle]

I'm currently shooting with the Canon S5-IS, and as with most of my settings, I set a custom white balance one time and used it ever since. Set up the lighting, fill the frame with a white card, and let the camera select a custom white balance for that setup. You shouldn't have to worry about it ever again, unless you change lights.

[Misty]

That generally does work fairly well, but I find it's not 100% consistent - like when white balancing by clicking on a white pixel in a raw editor, the results are usually close but off to some degree. I suppose it depends on how important colour reproduction is to you - I'm working with many archival documents, where I want to have very close reproduction. For that kind of purpose, I'm finding that tuning to the temperature of the light is giving more precise results, but not everyone might need that precision.