Why white balancing in your camera, may be a bad idea.

http://www.youtube.com/watch?v=-Km_T3sR9fs

That's the video, and it was either set to auto or daylight, I believe. The T2i doesn't have the tempeture white balances, it just has the mood presets and to be honest, I don't really know which one of those creates the least noise. Anyway, in the video, just watch it full screen and you can see all the crap being harsh in the background blur, it's pretty awful to me and that's what it looks like on the native video. Couldn't ever use something like that in a film, which the shot was practice for a very similair shot I'm doing for an upcoming film of mine.

I just don't know why it really is that harsh though, I doubt it's my computer making it look that way on both the original file and on youtube, I mean it really, really looks very bad, it looks more like the noise levels you see when you zoom into a video still a good portion.
 
There has to be something wrong with the testing. Considering we are white balancing to achieve identical white point (same digital gains), it's impossible that white balance in post will be better than white balancing in camera. Doing it in post will always quantise more and boost compression artifacts. When we are dealing with a very high precision uncompressed recording, doing it in post will only match the results, it will not be better. Additionaly, white balanding produces a stronger luma which generally helps compression.

From better to worse:

1. Optical white balance with filters (if the camera signal processing is compatible of setting R,G,B gains individually to 1.0)

2. Optical white balance using the filter wheel (typically inferior because a tradeoff for sensitivity still forces some digital gain).

3. Digital balance in camera (always performed at full resolution before gamma, bit depth reduction and compression).

4. Digital balance in post

There is small possibility though that the camera compression is so bad that it needs a powerful chroma to capture it with detail, in this case, in-camera white balance will lower one chroma channel and compress it with greater artifacting. This has a tradeoff in luma and the other chroma channel though. I doubt it will ever provide an advantage.
 
White balance in post is also problematic because of color shift. Proper white balance is performed in linear light. Doing it on a "Rec709" gamma will distort the chroma space. An advanced app could linearise the file, perform white balance and go back to video gamma, but that will not work perfectly well because modern cameras have very unique gamma curves, unknown to the software manufacturer.

Also, white balancing in camera (where the manufacturer has control) or in post with tools provided by the camera manufacturer (such as RAW converters) can allow a superior noise look. But I don't want to say more on this:)
 
Assume a 10stop ASA400 single sensor camera at 3200K recording 12bit linear uncompressed RGB files. The gain required for 3200K in this camera is rgb=(1.0,1.0,2.0). Blue is half of what it should be.

1) We can use a prescribed blue filter. We get maximum dynamic range, lowest noise on all three channels. We have lost a stop in sensitivity. We need to compensate with light.

ASA200, (10,10,10) stops, (12,12,12) bit precision, no color shift
ASA400, (10,10,10) stops, (12,12,12) bit precision, no color shift, with the additional light.

2) We can do digital white balance in camera. We get 1 stop lower dynamic range in Blue and keep the ASA rating.

ASA400, (10,10,9) stops, (12,12,11) bit precision, no color shift

3) We can do it in post. Exactly the same with the previous scenario since we have all the original precision to work with. Asumming linear 12bit, we don't get color shift.

ASA400, (10,10,9) stops, (12,12,11) bit precision, no color shift

4) Same with (2) but lowering the gains of red and green instead of increasing blue gain. We get a 50% maximum level which we need to compensate for in post. This wastes 1 stop and 1 bit in all channels. There is no advantage vs (3).

ASA400, (9,9,9) stops, (11,11,11) bit precision, no color shift

With compression and video gamma applied it's more complex because there are color shifts.

There is the possibility than analog gain is used in a 3CCD camera for white balance and the camera analog CCD support electronics are so poor that analog white balance gain produces results inferior to digital gain. The test was done on an HVX which is 3CCD. This could explain the result in Frank's test. But why didn't they use digital white balance in camera if the analog gain is inferior? Digital gain would be superior to doing it in post.
 
You must log in or register to reply here.
Back
Top