Exposure compensation

[vorlich]

1. If your in-camera histogram is bunched up at the extreme right, then your shot is probably worthless anyway.

2. The default histogram in the raw processor will be similar to that displayed in-camera if you are using a raw processor, which applies the setting; the camera makers' own raw processors do, but for example ACR does not (except for the WB).

However, that is again the histogram of the converted image, which is far from the histogram of the raw data.

1. Thanks for those pearls of wisdom. :-/
2. I'm talking LR here. All settings zeroed after import. I see a histogram broadly similar to the in-camera histogram.

[Panopeeper]

2. I'm talking LR here. All settings zeroed after import. I see a histogram broadly similar to the in-camera histogram.

Ok, pls upload a sample raw file of yours for demonstration (I have several thousands from five dozen camera models beside the tens of thousands of my own, but a demo is best on your own file). Yousendit.com is the best way.

[digitaldog]

1. Thanks for those pearls of wisdom. :-/
2. I'm talking LR here. All settings zeroed after import. I see a histogram broadly similar to the in-camera histogram.

Because you're viewing a gamma corrected, rendered histogram. Again, the current histogram is based on the current  condition of the data. What you're now seeing isn't the Raw data of course. And those rendering settings may not be anything appropriate for the color and tone appearance you'll end up with.

Make an exposure that is 1 stop brighter than the meter recommends. The LCD should show its over exposed and clipped. The Histogram of the default rendering in LR will show the same. Now use the Exposure slider and move it down minus about 1 stop. You should see a preferable color and tone appearance and histogram. But wait, if the image were really over exposed, you couldn't do that. The image wasn't over exposed, at least as for as the Raw data was concerned. It was for the in camera JPEG generation. So the Raw histogram is what it is. Exposure and ISO affect it this of course. At some point, you really will over expose to the point that you've saturated the sensors and blown out the highlights. But you don't know where that is just shy of happening based on the JPEG generated histogram.

In the URL I provide, I was able to "over expose" (some would say properly expose) 1.5 stops over the meter recommendation while ending up with a final image that wasn't over exposed. So how do we attribute this? The Raw data if visible as a histogram would have shown me that 2 stops indeed clip, 1.5 doesn't. But the histogram on the LCD told me both were over exposed. So which is it? Well for JPEG, that histogram is appropriate because it defines what the camera would hand me based on its rendering. Its not appropriate based on what I ended up doing in Lightroom by adjusting the exposure slider.

[vorlich]

Ok, pls upload a sample raw file of yours for demonstration (I have several thousands from five dozen camera models beside the tens of thousands of my own, but a demo is best on your own file). Yousendit.com is the best way.

Why don't you illustrate your point with one of your own files?

[bjanes]

I don't recall Bruce saying it would not be useful, can you find a reference? Yes, it would be all scrunched up to the left, especially if plotted like a gamma corrected Histogram. And yes, users would need to learn to read a histogram that way. And yes, the manufacturers could do something useful like put some line or reference to the right of this scrunched up histogram that indicated based on some testing, where full sensor exposure overload happened so we could adjust to just shy of that point. And yes, that would be real useful IMHO!

He made this statement in response to one of my posts (?Adobe Forums or LL forums), where he was trying to discount my advocacy of a camera raw histogram.

Accurate by how much? You have some data to share? 1/10 a stop? 1/2? I've fooled around with my Canon's and 1.5 stops is more like it. Now maybe you consider that an accurate amount to extrapolate but I'd like a bit more precision.

Accurate to about 1/3 f/stop, which is adequate for most practical work. Most camera settings and light meters to not offer finer settings.

To illustrate and prove my assertion here are photos of the D3 histograms for my images 005 and 006 which are shot at 1/20 f/8 and 1/25 f/8 respectively. The camera histogram of the image at 1/20 sec shows clipping of the highlights, while the histogram of the image at 1/25 shows the highlights slightly short of clipping.

[attachment=13466:005_cam_histogram.png]

[attachment=13467:006_cam_histogram.png]


And here are the raw histograms as shown by Rawnalize. The raw histogram of the image taken at 1/20 sec is very slightly short of clipping, while the raw histogram of the image taken at 1/25 sec is about 1/3 stop short of clipping.

[attachment=13468:005_raw_histogram.png]

[attachment=13469:006_raw_histogram.png]

I posted mine here:

http://www.digitalphotopro.com/technique/c...ng-for-raw.html

Maybe Canon's are way, way different from Nikon.

Maybe so, but you didn't look at the raw file directly, but used Adobe Camera raw which uses an exposure offset and doesn't really give you a good idea of the actual content of the raw file. To see the actual values in the raw file it is best to use specialized software such as Rawnalize or Iris. When you get some more data, we can discuss the matter further.

Bill

[Guillermo Luijk]

I agree with Bill that a totally linear histogram would not be so useful as a gamma or EV histogram. Applying a gamma to the encoded data, or plot the X-axis on a log 2 basis is no problem.
The gamma will never blow anything that is not blown in the RAW file and will help visualize better how the information spreads along. The EV divisions can also be plotted in a gamma histogram.
However an EV histogram would be my preferred option, it's conceptually very photographic and allows to find out at a glance the dynamic range of the scene.

Pick the one you like best:





Regards

[Panopeeper]

Why don't you illustrate your point with one of your own files?

I can show you the raw histogram of many files, but the point is, that you compare it with the histogram of the very same file in-camera and in your choice of raw processor with your settings.

However, do not mix up thing, I was not asking for a favour but offering one, for you don't understand the issues.

[digitaldog]

Maybe so, but you didn't look at the raw file directly, but used Adobe Camera raw which uses an exposure offset and doesn't really give you a good idea of the actual content of the raw file. To see the actual values in the raw file it is best to use specialized software such as Rawnalize or Iris. When you get some more data, we can discuss the matter further.

Yes absolutely true. I don't have anything for the Mac (at least I know of nothing) that allows me to look at a Raw histogram. But I did find that 1.5 stops was about the limit over the meter recommendation I could expose to the right for use with that converter. Which begs the question, was the 2 stop really the limit or did I hit an LR/ACR limit (going back to your point that you really want to view that Raw data).

[bjanes]

Yes absolutely true. I don't have anything for the Mac (at least I know of nothing) that allows me to look at a Raw histogram.

DCRaw works with the mac. You can demosaic to a 16 bit TIFF without white balance and then use Photoshop or some other program to view the histogram.

Here is the result for my 005.nef file whose raw histogram was shown above and repeated here for comparison. I used the following switches in DCRaw: -4 -T -o 0 -r 1 1 1 1. Perhaps Gabor and Guillermo can comment as they are more familiar with these programs.  (Note: If anyone notes a change in the histogram, I improved the originally posted Photoshop histogram by cropping out black pixels):


[attachment=13486:PS_histo.png]

[attachment=13472:005_raw_histogram.png]

If you want raw pixel values, you can read them with the eyedropper using 16 bit mode and then divide by two to convert to 14 bit raw values.

For example, here are readings in Photoshop and Rawnalize.
[attachment=13474:PS_info.png]

[attachment=13475:Rawnalize.png]

The raw pixel value of the indicated area is 15908 in Rawnalize and 31384/2 = 15692 in the Photoshop derived method.

[Guillermo Luijk]

I used the following switches in DCRaw: -4 -T -o 0 -r 1 1 1 1. Perhaps Gabor and Guillermo can comment as they are more familiar with these programs.

That's fine to find out what the RAW histogram looks like (in fact it's what I did 2 posts ago), but it's actually not pure RAW since you are demosaicing (without white balance nor output profile conversion, that's true).

For real RAW histogram you can do:
dcraw -v -D -4 -T -t 0
dcraw -v -d -r 1 1 1 1 -4 -T -t 0

The first is pure RAW data, just decompressed when needed. It's very useful to check fake ISOs and work out RAW saturation points.
The second is 'nearly' RAW data (it clips black point, adjusts saturation point and converts to 16-bit scale). It's very useful to obtain log 2 RAW histograms.

None of them performs demosaicing and can be represented in Histogrammar's RAW mode. Histogrammar has no problem to run on a Mac with Darwine or similar.

BR

[bjanes]

I agree with Bill that a totally linear histogram would not be so useful as a gamma or EV histogram. Applying a gamma to the encoded data, or plot the X-axis on a log 2 basis is no problem.
The gamma will never blow anything that is not blown in the RAW file and will help visualize better how the information spreads along. The EV divisions can also be plotted in a gamma histogram.
However an EV histogram would be my preferred option, it's conceptually very photographic and allows to find out at a glance the dynamic range of the scene.

Pick the one you like best:

Regards

Guillermo,

IMHO, the EV histogram is by far the preferable one.

While we are talking about camera histograms, I should mention that most camera histograms are luminance histograms and may not detect clipping in the blue and red channels. The camera RGB histograms will detect clipping, but are liable to clipping on white balance and will not give a reasonably accurate picture of the raw channels unless UNIWB has been loaded into the camera.

Bill

[digitaldog]

DCRaw works with the mac. You can demosaic to a 16 bit TIFF without white balance and then use Photoshop or some other program to view the histogram.

Cool, thanks, I'm on it.

[Panopeeper]

The in-camera histogram can be made very raw-like, if the camera allows for loading a user-defined curve. My Canon 40D supports this, but unfortunately only with a very primitive curve: I can define only one single point of the curve. I tried to create an "RGB inversion" curve, and this in fact enhanced the displayed histogram (causing the preview become darker), i.e. it is now closer to the raw histogram.

If other cameras support more complex curve specifications, the result can become much better; it should be even possible to achieve a log2 based display.

Note, that the flashing clipping indication does not change by this curve.

ADDED

I am an airhead. I CAN create a curve determined by up to eight points, only that the editor is a bit clumsy, one had to click very accurately on the curve.

[stamper]

In the URL I provide, I was able to "over expose" (some would say properly expose) 1.5 stops over the meter recommendation while ending up with a final image that wasn't over exposed. So how do we attribute this? The Raw data if visible as a histogram would have shown me that 2 stops indeed clip, 1.5 doesn't. But the histogram on the LCD told me both were over exposed. So which is it? Well for JPEG, that histogram is appropriate because it defines what the camera would hand me based on its rendering. Its not appropriate based on what I ended up doing in Lightroom by adjusting the exposure slider.
[/quote]

When you are making your exposure are you making an exposure based on the "mid tones"? Not for the highlights or dark areas? I often make an exposure for the sky - in landscapes - and set an EV of + 1 or expose for a dark area and set an EV of -1 How would this method fit into the equation? TIA

[francois]

Cool, thanks, I'm on it.

Rawnalyze and Iris do work on Macs (Intel only) with CrossOver Mac. No need to install Parallels or Fusion with Windows. CrossOver Mac is available as a demo ( http://www.codeweavers.com/products/cxmac/ ). I haven't been able to run Guillermo's software in CrossOver...

[EdRosch]

Hi All,

I'm confused about exposure compensation. I understand that the metering in most cameras will try to expose bright scenes as middle grey, snowy scenes for example. However, what I don't understand is the benefit of using exposure compensation over say, taking a shot, checking the histogram and adjusting the exposure manually by changing the shutter speed to taste. This seems much less fiddly than going into a menu and making the adjustment there.

Am I missing something?

Hi,  while the discussion about histograms is quite illuminating, it does seem a bit sideways to your question.  First, as a brief intro- the photo school I attended back in the film days taught previsualization and a variant of the Zone System.  They also required us to print on an ungraded paper (Ektalure G) and we were not allowed to use variable contrast papers or adjustable contrast paper while printing, thus our exposures had to be 'nuts on' or a high quality print would be impossible.

All that said, exposure compensation is about keeping focused   on the scene in front of you and not on your screen as well as developing an understanding of how your equipment is going to work with that scene.  That way meters, histograms and such become the tools they are meant to be and not the crutches they seem to have become.  

Now that said... exposure compensation is counter intuitive.  You have to stop down for dark scenes when you might think you'd want to open up and open up for light scenes when you'd think you'd want to stop down.  Note that this has nothing to do with the amount of light but with the overall tonality of the scene.  So if you meter snow in heavy overcast or bright direct sun you'd still have to compensate about the same.   Likewise for, say, dark lavarock under various lighting conditions.

So, the lesser advantage of exposure compensation is speed.  I shot a lot of snow and ice this past winter and it was a matter of moments to use the thumbwheel to dial in an offset.  The big advantage is that it encourages you to look more deeply into a scene than your might otherwise do and think about how it will interact with your specific photographic system in order to get exactly the results you wish.

Ed

[digitaldog]

When you are making your exposure are you making an exposure based on the "mid tones"? Not for the highlights or dark areas? I often make an exposure for the sky - in landscapes - and set an EV of + 1 or expose for a dark area and set an EV of -1 How would this method fit into the equation? TIA

The exposure were all based on an incident flash meter in these examples.

[digitaldog]

Rawnalyze and Iris do work on Macs (Intel only) with CrossOver Mac. No need to install Parallels or Fusion with Windows. CrossOver Mac is available as a demo ( http://www.codeweavers.com/products/cxmac/ ). I haven't been able to run Guillermo's software in CrossOver...

What's the difference, advantage of CrossOver versus Parallels (which I have on one machine)?

[Guillermo Luijk]

What's the difference, advantage of CrossOver versus Parallels (which I have on one machine)?

As far as I know Parallels (so as Vmware Fusion) are Windows virtual machines that require the installation of the Windows OS. Crossover (and Darwine) are Windows library emulators that allow to run Windows applications without the need to install Windows, and I guess lower degree of compatibility.

BTW francois, I have seen Histogrammar running on Crossover without any problem, maybe you have to do some config adjustment.

BR

[digitaldog]

As far as I know Parallels (so as Vmware Fusion) are Windows virtual machines that require the installation of the Windows OS. Crossover (and Darwine) are Windows library emulators that allow to run Windows applications without the need to install Windows, and I guess lower degree of compatibility.

Thanks, I'll check it out. Not having to actually install Windows sounds very handy!