[Turtle]

Hi,

I've been trying to use the ETTR (Expose to the right) method recenty, it is generally a good thing, but I'm finding (with my screen) that if there is bright oranges/yellows, that they seem to clip far more quickly than other colours.

Is this a sensor specific thing, or is this typical of digital photography?

Here is a shot showing the example, the candidate was exposed to the right (with no clipping, from highlights or reds), yet it seems to have lost all tonality.


Regards,

[dms]

In some rough tests I found--using tungsten lighting, K20d, and LL60 screen (which I seem to recall people saying meters about +2/3 or +1 stop)--the following increased exposure causes my raw file (in CS2) to be just blown!

red: +5 2/3 stop
blue: +5 stop
green: +4 2/3 stop
yellow: +4 2/3 stop
grey (white): + 4 1/3 stop

This is counter to what you find--but anyway it is reasonable that it should depend on color (also b/c bayer pattern has twice as many green sensors). As I said my testing was rough and if someone has more definitive test/reference it would be interesting.

BTW one great thing about digital (for me) is I can do these kinds of tests easily. I actually had a cheat sheet stuck on my camera--of exposure bias based on color I was metering--but I ran out of places to put things, and hyperfocal distance vs focal length and f stop, and the like, were more important for me.

[hcarvalhoalves]

One thing you want to make sure is that you're shooting in RAW with neutral color balance (no "vibrant", etc.).

That said, I've always noticed DSLRs have trouble reproducing brightly lit reds and oranges, but I'm not sure if it's the sensor fault, the lenses, or combination of both.

What I do know, though, is that my K-x renders more natural tones when paired with a Pentax-A 35-105mm or 50mm than with the kit DA 18-55mm, which seems to produce warmer colors and often clips red highlights, specially under sunlight (which is, theoretically, pure white). It might be some difference in the coatings, or maybe the camera processing engine attempting to correct for DA lenses, but I've never identified where the problem lies so I just stick to my A lenses instead.

[dms]

My apologies for not being more careful--you said "orange and yellow" and I was thinking red. My results suggest yellow will clip earlier than other colors. (But not earlier than white! Which seems a bit odd/inconsistent?)

Also to be clearer: my results are using the K20D spot meter to obtain exposure for a particular color swatch, and then pushing the exposure till it just clips in ACR CS2.

[feilb]

Hi turtle,

First I'll ask: Which camera are you using?

If you are using the K5, it may not be in your best interest to do so depending on the situation. That is a whole other issue

Now onto your post. The histogram (which is displaying the values for the image on the right) does not indicate any highlight clipping whatsoever. If there was clipping, you would see a big spike bunched up at the very right hand side of the image. In fact, you could have gone a little further to the right without losing information!

That being said there are a couple of things to note:

1. Most monitors, especially un-calibrated monitors, do a poor job of displaying the full range of color values, especially at the extremes of light and dark. This is especially true of the panel technology used in most inexpensive LCD monitors: TN. You can make an attempt at fixing that by going through the calibration procedure at this site: Monitor Calibration.

2. In general red is one of the first colors to clip on a typical sensor because many CCDs are more sensitive to red light than to blue light. This can be a challenge to deal with in scenes with wide dynamic range (very light and very dark in the same image) because in order to keep red from clipping, you under expose the rest of the image. For cameras prone to this type of issue, multiple exposure combination (sometimes called HDR) can be your friend. If that is not an option, i would recommend exposing a bit to the left and compensating with fill light.

[feilb]

My apologies for not being more careful--you said "orange and yellow" and I was thinking red. My results suggest yellow will clip earlier than other colors. (But not earlier than white! Which seems a bit odd/inconsistent?)

Also to be clearer: my results are using the K20D spot meter to obtain exposure for a particular color swatch, and then pushing the exposure till it just clips in ACR CS2.

dms,

Your metering methodology may be masking the actual behavior of the sensor as the meter is likely not equally sensitive to all colors.

A better test would be to perform the same test under the exact same lighting and exposure conditions, then repeat your post processing test.

[dms]

feilb--thanks. You are correct. My test/interest was spot metering the brightest SPOT of clothing under uneven lighting, and getting IT to just avoid clipping. This is a combination of meter and sensor. This is not the typical scenario using weighted average or matrix metering--but if using ETTR and camera spot metering the scene--I think it may be applicable.

[Turtle]

Thanks all-
Felib: I am using a k5, so what is it in particular you are eluding to? I agree with the histogram, but as you can see, there is tonality loss already. My monitor is good - not great - and usually is fine. can you see details in the highlight on the candidate?


hcarvalhoalves: yes, I always shoot raw, and am looking to get my exposure technique correct as I post process all my images worth keeping (which are few).

Any more ideas?

[feilb]

Turtle, I do calibrate my monitor and can see detail in the highlight areas in the image you posted. Have a look at this site and see if you can see differentiation in the white tones on the right.

[dms]

Turtle, if your usual (not ETTR) exposure looks good (and prints well) then ETTR should look too bright/washed out! Cranking down the exposure will give you the look you like, and more detail/less noise in the shadows. ETTR normally means more processing work on your part.

[Turtle]

Felib: Thanks for the link, and yes I needed to calibrate my gamma, and things look better.
DMS: Yeah normal exposure is great, but I'll continue to try and use ETTR, as I'm happy to reduce exposure after etc.

Now looking at previous exposures, there is less of the phenomenon, yet it's still there. I'll experiment as I shoot live performance and lighting is usually warm, so maybe I'll have to cool the white balance down to reduce red clipping.

Thanks heaps.

[feilb]

Turtle,

The reason I asked whether you were using a K5 is because, in my experience, the k5 has more latitude on the bottom end than on the top.

Exposing to the right is a technique used to attempt to get the most usable information out of the sensor by ensuring that the signal to noise ratio is high. Exposing to the right adds more signal to make the ratio more favorable. The issue comes in when the scene has high dynamic range. Because the read noise of the k5 is so low, I've found it beneficial to expose to the left so as not to blow any highlights, then to pull detail from the shadows ( believe me, there's a good two or three stops down there that most digital cameras can't use).

I hope I explained that ok.

[Turtle]

Thanks Felib, I'll have a play with ETTL on the K5.
I've mucked around with fine-tuning the images again and have got two of them to where I feel they are relatively accurate to life (but still not there).
The select was ETTL and the candidate ETTR, I think the ETTR one was harder to retrieve the highlights and get tonal balance. Neither was exposed correctly.

I'm not sure which one looks more natural.

This has been a very challenging experience, I think I'll use Oranges as a means to develop (pun) my processing and exposure skill.

Cheers.

[Malo1961]

A very informative thread about LR, Colourshifts and ETTR is found on the LULA forum. It will provide you with some valuable answers. If you haven't found it already, that is.