I ran two tests to address some misconceptions about ETTR.

The first test was last night just after 6PM. I used an A7 II that is converted to monochrome. RawDigger displays only a luminance histogram because there are no color channels.


1/5000s @ f/8 ISO 400


1/4000s @ f/8 ISO 400

The first image has about 187k raw values blown out. Since the sun was too close to the frame I aimed a little more to the left and was able to increase the exposure from 1/5000s to 1/4000s. The second exposure had only 2k raw values blown out.

The Zebra stripes make it easy to maximize the exposure without blowing many raw highlights.

The image covers a 10 stop range so the result SOOC looks very dark:



Nevertheless, the shadows can be brightened to produce an acceptable image:



This example proves that the notion that an image exposed using ETTR will always look blown out.

For the second test I used a Df at around 9:26AM.


1/1600s @ f/8 ISO 400 (Sunny 16, also the exposure recommended by matrix metering) - no blown raw highlights

I was seeing no highlight warnings so I increased the exposure by one stop:


1/800s @ f/8 ISO 400 - about 1100 blown raw highlights, not enough to matter

After processing both images and converting them to B&W I ended up with:


Sunny 16/Matrix metering exposure


One stop brighter

Development consisted a little shadow recovery and B&W conversion removing some the blue and cyan. I could not remove more because the shady side of the building was lit by only the blue skylight.

What this test shows is that a one stop increase in exposure does not produce a visible improvement in the final image. It also shows that Sunny 16 and matrix metering are safe and sufficient for daylight images. And finally, it shows that the JPEG SOOC for the ETTR image, although just past the raw limit, does not look washed out:


JPEG SOOC

Not at all a silly question.

Here is the example from page 5 presented a little differently.



The histogram Y-Axis range



makes it easier to see the ends of the plots and the number of pixels where the count is low. The raw value scale is from 0 to 16383 and raw values double at each one-stop increment. Most 14-bit cameras do not normally reach 16383 but they will come within about 0.05 stops of that limit.

The spikes on the right also make it clear that the green and blue channels are at their limit. The comb-like appearance on the left ends of the plots also give us a hint that there may be loss of smooth tonality below a raw value of about 64.

The details in the upper center give us a count of the number of pixels that are blown out in each channel to tell us more about the green and blue spikes. They also indicate that about half of the 16MP image is in the 0-64 raw value range.



RawDigger is an excellent inexpensive tool for researching the relationships between the raw data and the JPEG. It is also a great way to screen your images and decided whether or not there is a potential for recovering highlight and shadow information.

This image was just for illustration. It points out that the chances of making a decent image out of this single capture are slim. To get a better result a second image with a couple of stops more exposure could be combined with it to reduce the appearance of noise that was brought out of the shadows. That might also call for a tripod and a lower ISO but this camera is already at base ISO 200.

So what I guess is confusing me is that in the graphs, both the Y & X axis are unlabeled, and that was giving me a brain cramp. I use the histogram in the camera a lot so thinking about it that way I do understand that to the left side of the histogram if the pixels bump up against the top limit of the Y axis, that indicates lost info (black info) and similarly if the pixels to the right bump up against the top limit of the Y axis, you have lost highlight info. I also understand that this histogram reflects the JPEG (as opposed to the RAW file). I believe the Nikon RAW file is 12 bits so that would be 2^12=4096 levels.

So clarify for me if you will what are the labels for Primary Y axis and the Primary X axis on your graphs (there also seems to be a secondary X axis in "EVO")? I think (or so it seems) the primary X axis is the "bit levels" which doesn't translate easy in my brain? I would have otherwise thought the Y axis would be the "bit levels", which in your case I guess are 16 bit (2^16=65,536 levels) and I would have thought the Y axis would be the "luminescence" going left to right from dark to white. So, in that case, if the pixels at any point along the X axis vertically exceeded the maximum "bit level", you could interpret that as lost information. But it doesn't seem to be that way. Although I could see that the primary Y axis is the bit levels and then I guess I don't understand what the Y axis is? How am I mis-thinking this?