Odd ETTL-II/histogram observations

[tim]

I was doing a little practice this evening before a shoot tomorrow evening, and I noticed an odd thing with the histogram (details below). If anyone has any idea about what's going on here, i'd love to hear your ideas. All images were taken with the 20D hand held, Tamron 28-75 at 30-50mm, 550EX, partial metering (which should be irrelevant if my understanding is correct), FEC set as described, no EC on the flash, and CF-14 sets ETTL-II to evaluative.

The issue is that the histogram on the camera appears to be inaccurate, in that it under-reports the exposure. The histograms on the camera and in DPP look the same to my eye, and don't agree with the histogram in CS2 camera RAW. The camera under-reports the exposure, in my opinion, which would cause you to overexpose the images if you work purely from the camera histogram. Take a look at the differences yourselves below, all are of the same scene, which is the same as the 2nd picture in the 2nd group (marked CS2 add EC+1).

By my visual observation of the image, and by looking at the clipped pixels in CS2, I believe CS2 is far more accurate.
The first (FEC0) is clearly underexposed by both graphs.
The 2nd (FEC +2/3) is underexposed according to the camera, but falls just short of being perfectly exposed in CS2.
The 3rd (FEC+1 1/3)is just a bit short of being properly exposed according to the camera, according to CS2 the red channel is limiting quite badly.
The 4th (FEC+2) is properly exposed according to my reading of the camera histogram, and WAY overexposed according to CS2.


http://mrwild.co.nz/unprotected/potn/20D-LS2-FEC0.gif http://mrwild.co.nz/unprotected/potn/20D-LS2-FEC2_3.gif http://mrwild.co.nz/unprotected/potn/20D-LS2-FEC1_1_3.gif http://mrwild.co.nz/unprotected/potn/20D-LS2-FEC2.gif

What's going on here?! Why does the camera histogram appear to under-report exposure?


The second thing I noticed is an observation that I thought i'd share, not an issue. It's that the background color makes quite a big difference to the amount of exposure compensation you need, which is just as you'd expect, really - it's evaluative metering. Kinda wish I had partial metering for flash now. Initially these results confused me, but when I thought in terms of how much light the background reflects back it made sense.

In CS2 Bridge you can see clearly that the first image (white background) is massively underexposed, the 2nd (mixed, mostly light colored background) is underexposed but not too badly, the 3rd (red background) is about right , and the 4th (no background - taken outside) is a little underexposed. The screenshots of the CS2 histogram was taken before EC was applied. The person was about 1 meter from the background for all the shots. I don't really understand why the black shot came out only needing EC+0.5, but I can just remember that bit.

http://mrwild.co.nz/unprotected/potn/WhiteBgd.jpg http://mrwild.co.nz/unprotected/potn/MixedBgd.jpg
http://mrwild.co.nz/unprotected/potn/RedBgd.jpg http://mrwild.co.nz/unprotected/potn/BlackBgd.jpg

[PacAce]

The histogram you are looking at on the camera LCD and DPP are luminance histograms. What you are looking at in PSCS2 ACR is the RGB histogram. If you want to know what the difference is between the two, there was an extensive discussion on it here:

http://www.photography-on-the.net/forum/showthread.php?t=70397&page=3&pp=20&highlight=green+channel

[tim]

Thanks Leo. I guess then I just have to learn how to read luminence histograms. It does suprise me that the image is clearly overexposed and the histogram doesn't clearly show this, or even give you a hint of it. I'll read that luminous landscape article again tomorrow, it's 1am here and I should be in bed.

[slin100]

This illustrates the key disadvantage of a luminance-based histogram. It can look fine yet one or more channels can be blown.

[tim]

This explains why the wedding photos I took a couple of weeks ago came out with the red channel blown, but the histogram not telling me anything was wrong. What's the solution? Is there one?

[PacAce]

This explains why the wedding photos I took a couple of weeks ago came out with the red channel blown, but the histogram not telling me anything was wrong. What's the solution? Is there one?
Yes, there is at least one solution that I know of (applied it myself, AAMOF) but I'm not sure if you're going to like it. It's called the 1DmkII, which can display a separate histogram for each of the three colors. :mrgreen: :lol:

[tim]

I plan to buy the 1D Mk 3 when it comes out, i'll not bother with the Mk II.