Problem with highlights with ETTR?

[Sareesh Sudhakaran]

I did my own crude test for ETTR and it confirmed what Michael said about low noise.

However, it did bring about one oddity: worse specular highlight detail. At the time I wrote my blog, I assumed it must be due to my poor lens. However, I just wanted to make certain if this was the case.

Here is a link to the test: http://sareesh.com/2011/08/ettr-expose-to-the-right/

The image in question is the last comparison, with the highlight on the cup. Do your experiences show any such behavior, or was it a problem with my test methodology or equipment?

[pegelli]

I think your test is ETFTTR (Exposed Too Far To The Right) and you blew more bright areas around the specular highlights.

ETTR only works if you don't (or don't mind) blowing more highlights. In the case of your test the specular highlights in the final shot are still quite small so could still be acceptable.

Bottom line, in higher contrast scenes you'll have to balance your desire for lower shadow noise vs. you desire to keep highlights in check.

Btw, the comment about Michael in the first comment on your blog is clearly showing that guy didn't read (or didn't understand) the article properly.

[Sareesh Sudhakaran]

It didn't clip the histogram. I checked it very carefully. But still, there might be something wrong with my methodology.

[pegelli]

It didn't clip the histogram. I checked it very carefully. But still, there might be something wrong with my methodology.

Probably the extra clipped pixels in the specular highlights are too few to show significantly in the histogram..
I think there are clipped pixels in both exposures in the specular highlights

[Sareesh Sudhakaran]

You might be right.

Other than the histogram, how else will we know if the image was clipped or not? Even during the shot, on the LCD screen, the highlights didn't blink due to over exposure. Neither did the histogram in DPP show as being clipped. Any other way?

[pegelli]

Only way I could think of would be if the "highlight blinkies" would also work in zoom mode, but I know on my camera it doesn't.

On the other hand if the specular highlights are so small does it really matter that much for your final picture? My own take on ETTR has been that it's a concept, not a mathematical exact way to determine the absolute correct exposure.
But realise that I'm usually not worried by a little bit of noise. First because noise shows much less in prints than looking at a 1:1 crop on screen and secondly because the denoise slider in Lightroom is so damn good.

[fike]

You have hit on a substantial limitation of EttR (which Michael does mention)--if you have an image with a wide dynamic range that will use up your sensor's full range from black to white, then EttR is of little value unless you know you want to let the highlights clip. 

To my mind, EttR is only really helpful in flat uncontrasty light.  I generally think of a foggy or rainy day photo where the entire histogram is bunched up in the middle of the histogram. In THAT case, EttR can make a mediocre exposure into a good one. 

Remember that the histogram is a pretty coarse measurement and that it is based on the JPG rendering of your image.  For these reasons, you can't be absolutely comfortable that what you see on the histogram includes all the data. 

To evaluate whether EttR will work for your scene, you need to evaluate the scene with your eyes and make an educated guess about whether or not there is data that is out beyond the right side of the histogram. 

[Sareesh Sudhakaran]

You're right. Those highlights show it wasn't a true low contrast scene.

[madmanchan]

Yes, the highlights most likely clipped in the raw data.  And if they clip, they're gone ...

[Peter_DL]

>> Shot RAW, processed in DPP: Picture Style: Neutral. White Balance: Shade <<

However, it did bring about one oddity: worse specular highlight detail.

Could be DPP-specific.

We have recently seen with another example that DPP likes to push areas of single RAW-clipped channels to pure white (even at Linear setting). Which could apply here to the surrounding of the truly blown, all-Raw-clipped part of the highlight spots in your test image, so that the result may appear like blooming.
ACR/LR were behaving differently.

Peter

--

[Daniel Browning]

And if they clip, they're gone ...

• If you ever clip your highlights of a river of molten lava, let'em go, because, man, they're gone.
• As the evening sky faded from a salmon color to a sort of clipped highlights gray, I thought back to the salmon I caught that morning, and how gray he was, and how I named him Flint.
• Photographers need encouragement. If a photographer gets an exposure right without clipping highlights, tell him it was a lucky guess. That way he develops a good, lucky feeling.
• When the highlights get clipped, it takes a big man to cry. But it takes a bigger man to laugh at that man.
• It's easy to sit there and say you'd like to have more highlight headroom. And I guess that's what I like about it. It's easy. Just sitting there, rocking back and forth, wanting those highlights.
• Laurie got offended that I used the word "puke." But to me, that's what her clipped highlights looked like.
• Maybe in order to understand highlights we have to look at that word itself. HIGHLIGHTS. Basically, it's made up of two separate words "highl" and "ights" What do these words mean? It's a mystery and that's why so are highlights.
• Sometimes I think you have to march right in and demand your highlights, even if you don't know what your highlights are, or who the person is you're talking to. Then on the way out, slam the door.
• To me, it's a good idea to always carry two sacks of something when you walk around. That way, if anybody says, "Hey, can you give me a hand restoring these saturated highlights?", you can say, "Sorry, got these sacks."
• Whenever someone asks me how to restore clipped highlights, I usually think for a minute, then I spin around and pin the guy's arm behind his back. NOW who's asking the questions?

(Apologies to Jack Handey.)

[mkravit]

What would Ansel say? 

[torger]

What would Ansel say? 

Ansel worked with film which behaves better than digital in highlights, instead of immediately clipping there's compression. With digital you have the RGB channels and they rarely clip at the same time, so you drop one color channel at a time (usually green first and red last), thus the RAW software must be smart to avoid color artifacts in clipped highlights. In some RAW software you can choose between different highlight reconstruction algorithms, some monochrome, some color-guessing etc, and it depends on the material which algorithm that will work the best.

If your digital camera has great dynamic range you'd never want to clip anything. In practice the cameras don't have that great range though, and keeping small unimportant highlights may lead to that you sacrifice quality of darker parts of the image, which for that particular image may be much more important than some small glary detail. I don't think it is wise to have as a rigid principle to never clip anything. In many situations highlights are monochrome and then you can clip all but one channel and still have relevant detail.

In still life photo where timing is less important, I usually bracket and decide in post-processing in the RAW developer what exposure is the best. With experience one learns where it works to clip some and when not.

[madmanchan]

This is why many cameras now tend to underexpose in-camera (i.e., meter low) and apply digital gain afterwards (in software): leave enough highlight headroom so that the highlights don't harshly clip, which looks bad. Some cameras even have a menu option for it (e.g., Canon's Highlight Tone Priority, Pentax's Dynamic Range Expansion, etc.)

[ErikKaffehr]

Hi,

There are highlights and highlights. Specular highlighst are normally not possible to reproduce. A typical specular highlight is a reflection of a intense light source. With ETTR we normally strive to expose so nonspecular highlights are correctly handled but ignore the specular ones.

Best regards
Erik

You're right. Those highlights show it wasn't a true low contrast scene.

[Sareesh Sudhakaran]

With ETTR we normally strive to expose so nonspecular highlights are correctly handled but ignore the specular ones.

Thanks. Just to understand you correctly: Are you saying that the specular highlights in my comparison are different because a specular highlight is unrepeatable, even when using the same instrument under controlled testing?

If yes, then does the worse specular highlight rendition in ETTR hold significance, or was it due to other factors like lens, sensor, DPP, etc.?

[madmanchan]

Sareesh, to generalize Erik's point a bit ... the important thing with ETTR is making sure that the stuff you care about doesn't clip (*).  If you're photographing shiny reflective objects, you may have very bright small areas in your image -- areas where the objects reflected bright light sources.  Or the light source itself (light bulb, candle, street lamp, etc.) may directly appear within the image.  Do you care about whether these very bright areas are clipped or not?  That's up to you. 

Eric

(*)  This is why histograms aren't a good solution for ETTR.  Not even raw-based histograms.  You may be able to tell from a histogram if something has clipped, but there is no way to infer from the histogram where in your image the clipping occurred (which is ultimately what matters).  This is why a raw-based clip indicator overlay on the image would be much more useful -- you could see a visualization of where the clipping occurred and decide for yourself whether or not you care.

[Bart_van_der_Wolf]

(*)  This is why histograms aren't a good solution for ETTR.  Not even raw-based histograms.  You may be able to tell from a histogram if something has clipped, but there is no way to infer from the histogram where in your image the clipping occurred (which is ultimately what matters).  This is why a raw-based clip indicator overlay on the image would be much more useful -- you could see a visualization of where the clipping occurred and decide for yourself whether or not you care.

Hi Eric,

While I agree with the preference for visualization of where the clipping occurs, the Raw based histogram could help in estimating how far below clipping the current exposure setting is. So, we'll probably need both.

Cheers,
Bart

[Peter_DL]

(*)  This is why histograms aren't a good solution for ETTR.  Not even raw-based histograms.  You may be able to tell from a histogram if something has clipped, but there is no way to infer from the histogram where in your image the clipping occurred (which is ultimately what matters).  This is why a raw-based clip indicator overlay on the image would be much more useful -- you could see a visualization of where the clipping occurred and decide for yourself whether or not you care.

OK, then, we want a fuzzy logic ETTR exposure mode, which rules the clipping percentage based on the analysis and experience with myriads of shots  .
Not sure where we are with today’s matrix metering.

Peter

--

[madmanchan]

I'm not opposed to a histogram but I think it's less directly useful to the photographer ...

... I think it's more useful to simply adjust the exposure up & down while looking at the (raw-based) clip indicators over the image.  Kind of like how it works in raw conversion software, but doing it in the camera instead at capture time.  That way you can just crank up the exposure till the clip warning covers something in the image that you care about.  Or going the other way, if there's a really important set of bright areas in your image, you can turn down the exposure till the clip warning in those areas go away (and use that as a starting point for bracketing exposure, for instance).