shooting to the right/digital blending

[dwdallam]

Which is why I recommend using a fixed white balance that matches the JPEG-based histogram to the RAW data as closely as possible. A RAW histogram won't require you to worry about WB while shooting, it will simply accurately inform you which channel(s) are about to blow out.
[a href=\"index.php?act=findpost&pid=86380\"][{POST_SNAPBACK}][/a]

Upon furter thinking about this, I think you are completely correct in this regard, with one caveat. The histogram must be able to accurately tell you when you are about to blow out highlights. If so, then you would want to expose to the right as much as possible, which would render your highlights as good as you can get, and your shadows as good as you can get. Then you can always under expose, if you need to--as when shooting high contrast scenese, such as a sunset--to bring out more color in the sky and clouds, and clip the shadows to black if necesary. Put another way, if you can get an accurate reading using your camera and histogram, expose to the right to retain as much dynamic information as you can.

[Jonathan Wienke]

Upon furter thinking about this, I think you are completely correct in this regard, with one caveat. The histogram must be able to accurately tell you when you are about to blow out highlights. If so, then you would want to expose to the right as much as possible, which would render your highlights as good as you can get, and your shadows as good as you can get. Then you can always under expose, if you need to--as when shooting high contrast scenese, such as a sunset--to bring out more color in the sky and clouds, and clip the shadows to black if necesary. Put another way, if you can get an accurate reading using your camera and histogram, expose to the right to retain as much dynamic information as you can.

That's the purpose of the test I described in my article; figuring out the exact exposure interval between the histogram indicating blown highlights and RAW clipping. As long as RAW isn't clipped, there is zero benefit to reducing exposure any further; you do not lose color saturation if you expose just below RAW clipping and then use a negative exposure setting in your RAW converter.

[Jonathan Wienke]

Yep, you are correct, WB isn't an issue. My bad.

Pobody's nerfect...

[John Sheehy]

That's the purpose of the test I described in my article; figuring out the exact exposure interval between the histogram indicating blown highlights and RAW clipping. As long as RAW isn't clipped, there is zero benefit to reducing exposure any further; you do not lose color saturation if you expose just below RAW clipping and then use a negative exposure setting in your RAW converter.
[a href=\"index.php?act=findpost&pid=86543\"][{POST_SNAPBACK}][/a]

This is generally true of high-key and normal-to-high-contrast subjects.  Converters don't always deal gracefully with exposure adjustments below -1 or so, though.  They don't seem to understand the concept of exposing to the right, at least for low-contrast subjects where you can go way to the right.  -4 exposure in ACR should theoretically render clipped RAW data at mid-grey or a little darker, but will render it at 255, anyway, stretching the extreme highlights like bubble-gum stuck to the ceiling.  An "Exposure" control, IMO, should just be a scaling of the RAW data, not a combination of that and a Curves tool, which is what ACR seems to do.

[TimothyFarrar]

Something I think you all will find very usefull, expecially for those who shoot action and therefor cannot blend exposures,

Back to the topic of White Balance (WB) and improving the Expose to the Right (ETTR) method.

The point of ETTR is to maximize the amount of the exposure which is in the most sensitive part of the camera's sensor without clipping the highlights to the point where the image is not salvageable.

There is a way to improve upon this --- apply the ETTR method to each color channel seperatly using a colored filter on the camera.

Ever notice that the noise levels in the shadows seem to be different for each channel?

We know that the RAW data is captured the same regardless of the WB setting in the camera. However we also know that the true WB of the scene does effect the ratio of light hitting the Red, Green, and Blue pixel elements in the camera.

Also, I suspect that there is another larger factor at play, the camera's Red, Green, and Blue pixel elements might have a different base sensitivity to light.

So I did a test. I took some RAW files (from the Canon 5D), one captured before dawn (blue white point) and one captured with indoor lighting at night (yellow/orange white point), and converted them using dcraw, which can be used to simply dump the raw linear data from the capture with no white balance adjustments (into a 16bit tif). Basically so I could see exactly what the camera was capturing.

What I found was that the output tifs have a very green tint to them regardless of the scene's true WB. Also I have noticed that with the Canon 5D most of the noise seems to be in the Red and Blue channels (try shooting with the 5D's long exposure noise reduction turned off). The data supports the theory that the camera is much more sensitive in the Greens.

When you use ETTR, at least with the 5D, it is simply the Greens that limit how far you can push your exposure to the right. The Reds and Blues will still be exposed to the left and hence have much more noise than the Greens. The RAW converter when adjusting the WB is simply scaling the Reds and the Blues to increase the intensity to match the Green channel to the proper WB.

Now if I had used a Magenta filter on the camera to drop the exposure level of the Greens, it would push the exposure of the Reds and Blues to the right as well, providing a much better capture in terms of noise (at the expense of a slightly longer exposure time or larger apature, and having to hand correct the WB later in the RAW conversion).

Each type of camera, not sharing the same type of CCD or CMOS sensor array, probably has a different base sensitivity to the RGB channels. So you could probably run through the same method I did, and figure out the perfect type of colored filter for a given range of scene WB, to maximize the ETTR method for all color channels and get a much lower noise image.

[feppe]

What I found was that the output tifs have a very green tint to them regardless of the scene's true WB. Also I have noticed that with the Canon 5D most of the noise seems to be in the Red and Blue channels (try shooting with the 5D's long exposure noise reduction turned off). The data supports the theory that the camera is much more sensitive in the Greens.
[a href=\"index.php?act=findpost&pid=86580\"][{POST_SNAPBACK}][/a]

A Bayer array filter found in almost all digital consumer cameras results in 50% green, 25% red and 25% blue pixels. These are combined by the RAW-processing app. This might be the reason for the  green tint in the minimally processed dcraw output, although it's impossible to say without knowing how dcraw does the processing. It also explains why digital cameras are more sensitive to green - it's not the sensor itself but the filter.

[John Sheehy]

A Bayer array filter found in almost all digital consumer cameras results in 50% green, 25% red and 25% blue pixels. These are combined by the RAW-processing app. This might be the reason for the  green tint in the minimally processed dcraw output, although it's impossible to say without knowing how dcraw does the processing. It also explains why digital cameras are more sensitive to green - it's not the sensor itself but the filter.
[a href=\"index.php?act=findpost&pid=86610\"][{POST_SNAPBACK}][/a]

The RAW data has a green or cyanish-green bias in most DSLRs, because the combination of the sensor and IR-cut and Bayer matrix filters causes a difference in sensitivity between the three color channels with full-spectrum white light.  The green is usually about a stop more sensitive than the red, and 1/4 to 1/2 stop more sensitive than the blue.  The fact that there are more green-filtered sensors has absolutely nothing to do with this.  Green was chosen to be doubled because it is most sensitive (and most light sources have plenty of green, but may be deficient in red or blue), and therefore the most light will be collected.  If there were 98% red pixels, 1% blue, and 1% green, the green channel would still be the most sensitive, and the RAW data would still have a cyan-ish green cast.  There would just be extremely poor resolution in the green and blue channels.

[bjanes]

Green was chosen to be doubled because it is most sensitive (and most light sources have plenty of green, but may be deficient in red or blue), and therefore the most light will be collected.  If there were 98% red pixels, 1% blue, and 1% green, the green channel would still be the most sensitive, and the RAW data would still have a cyan-ish green cast.  There would just be extremely poor resolution in the green and blue channels.
[a href=\"index.php?act=findpost&pid=86630\"][{POST_SNAPBACK}][/a]

Another reason for more green pixels is that the eye is most sensitive to green.

[John Sheehy]

Something I think you all will find very usefull, expecially for those who shoot action and therefor cannot blend exposures,

Action is the hardest place to use color filters, unfortunately, since you are often running on a tight photon budget, needing a fast shutter speed, and possibly also some DOF for sudden changes in subject distance.  If, however, you were already operating successfully at a lower ISO, you could go up an ISO and still get better results, I think, as the image is only as strong as the noisiest color channel.

Each type of camera, not sharing the same type of CCD or CMOS sensor array, probably has a different base sensitivity to the RGB channels. So you could probably run through the same method I did, and figure out the perfect type of colored filter for a given range of scene WB, to maximize the ETTR method for all color channels and get a much lower noise image.
[a href=\"index.php?act=findpost&pid=86580\"][{POST_SNAPBACK}][/a]

I've been aware of the magenta native WB for a few years now, and I've stopped holding my breath, waiting for filter companies to come out with digital versions of daylight and tungsten filters.  Falsehood propagates faster than truth, and one of the fastest propagating falsehoods is that "digital eliminates the need for color filters".  Most DSLRs have very similar native white balances, even Canon CMOS vs Nikon CCD (multiplying the red channel by 1.9 and the blue by 1.4 in the RAW data of almost any DSLR should get you close enough for exposure purposes; small tweaks in the converter can make up for any differences).  A single series of filters for most DSLRs should work well for many of them.

Again, filters don't help unless you can get a good exposure with them.  Better to have the red channel one stop from "the right" than to have all three channels one and a half stops from "the right" with a filter.