How to get the best raw exposure, part 2

Be careful of what you wish for 🙂

One of my wife's Mother's Day presents this past Sunday was some potted red flowers, I took their appearance as an unmistakable message from the forum gods that I should go ahead and do a red flower test of my EM1iii's live (luminance) histogram and highlight warning blinkies. Contrary to Donald's results with his magical Sony cameras, my test results confirm that my EM1iii's live histogram is based on the RGB data that generates in-camera display and JPEGs. The real difference between my results and Donald's has nothing to do with the different cameras used and everything to do with different scenes tested (and resulting highlights in the respective color channels). If Donald were to perform a similar test with his Sony cameras on a similar red flower scene as the one I used, his Sony cameras would no longer display luminance histograms that seemingly magically match the corresponding raw histograms. Here are my test results:

With 0 exposure comp. the live histogram (upper left) is pushed well to the left and looks like there's absolutely nothing approaching clipping. However, the post-shot in-camera RGB histograms (upper right) show that the red channel is pushed well to the right but just short of clipping. In this scenario, the live histogram does a dangerously poor job of predicting red channel behavior in an in-camera generated JPEG shot. And as can clearly be seen from the FastRawViewer screen grab, the live luminance histogram looks nothing like the raw histogram either, which in turn looks pretty different from the in-camera RGB histogram. The only good news here is that, at least, only a measly 3 red pixels were clipped. The bad news is that there's a significant amount of blocked/underexposed blue pixels in particular.

[]
(/a/ylJFCkDvVpWc9OGSJOURjPJsy6UfkvGEIycV3at8XQo48iRvBUfY8RQcQVtyBd6B/8566/?shva=1)

The above collection of screen grabs is for the same scene shot with +2 EC applied. Not surprisingly, the live histogram is pushed farther right but still shows plenty of breathing space from the right edge of the histogram. So, no clipping, right? Wrong for the JPEG and just barely wrong for the raw. The 5k clipped red pixels in raw can easily be addressed in conversion. However, the JPEG red channel clipping means that any attempt to pull back image lightness in postprocessing of the OOC JPEG will result in a color shift toward orange/yellow for the clipped red pixels. Note also how nicely the luminance histogram matches the green channel in the in-camera RGB histogram and also the green channel in the raw histogram...but ONLY the green channel. There's no way to argue that it matches either of the other raw channels or a combination of the three raw channels.

The above collection of screen grabs is for the same scene shot at +2.3 EC. Everything looks similar to the prior shot, as you'd expect from the mere .3 ev increase in exposure. I've included it here though because this is, for me, the best ETTR'd shot I took within the 3 stop bracket of 1/3 stop increments. Although 5.7% of the red raw pixels are clipped, they're speckled and scattered across the flower petals and not clumped together in a way that leads to visible color shift or that flat, textureless color clipping we always dread with red flower shots (viewed with a narrow gamut display/print). With careful conversion and maybe a bit of postprocessing trickery in Photoshop (e.g., channel blending), means that these pixels can be sacrificed for the noise and shadow improvements to be gained in the blue channel (which still has some blocked pixels). However, I suspect many ETTR shooters would not want to venture beyond the more conservative +2 EC shot. Different priorities, different workflows and skillsets...

The above collection of screen grabs is for the same scene shot at +3.0 EC. It's noteworthy for a couple of reasons. First and foremost, it still looks like the live histogram hasn't quite bumped up against the wall (likewise for the post-shot RGB luminance histogram). However, the first hint of a highlight warning blinkie is visible right in the middle of the white stigma. Things are also getting messy at the raw level. 30% of the red raw pixels are blown and that means there will be some significant color shift to the orange/yellow side of things in much of the flower. The green channel is also starting to get into the clipping game as well. This version is just living life too dangerously. Final note: still no good match between the live luminance histogram and the raw histogram. Of all of the individual channels, the best match to the live luminance histogram continues to be the green channel from the in-camera RGB histogram. Hmmmm...now what does that tell us about what is the primary driver for the shape and position of the live luminance histogram???

The above are quick dirty renderings of the 0 EC, +2.3EC and +3 EC shots done in ACR with roughly equalized lightness applied via highlight, shadow and exposure slider adjustments to illustrated the respective tradeoffs in shadows, highlights and color shifts.

I'll be out of pocket for several days, but I hope this might stimulate some legit discussion that gets the thread back on track. Unfortunately, I probably won't be able to reply to questions/comments until next Sunday.

your testing method is not going to be accurate because the red channel is not going to be the same colour red , i shot a red lunch box and my in-camera histogram was pretty well spot on .given the colour red is a mix of rgb colours. i might be wrong ,but someone more experienced might chip in.

You are presented with a real world result.

It's a good idea to study Bill Claff's dynamic range graphs too. I don't think your FF camera has to be made by Sony.
But it does seem that in general the larger format sensors have greater dynamic range.

Don Cox

your testing method is not going to be accurate because the red channel is not going to be the same colour red , i shot a red lunch box and my in-camera histogram was pretty well spot on .given the colour red is a mix of rgb colours. i might be wrong ,but someone more experienced might chip in.
[/quote]

Red plastic isn't a problem, unless it's fluorescent. Red flowers are the problem.

I think you should state what you think is the correct facts, but not get into a long argument with somebody who has his own collection of facts.

We each live in our own private universe.

Don

Single malt applied correctly gives you infinite patience.

You're a bad lad, Ted.

Don

I say to myself "What a maroon !" and leave it. Life's too short to worry about always winning arguments.

Don

I once read of a way to post something and then explicitly leave the last word to one's opponent, such that the opponent "lost" if they responded.

It's a great pity that I've forgotten the type of wording of that "something" ... a means to make a point and "win" either way ...

Yes, and this forum has effectively become “dpscience”. Which is fine if that’s the expected direction. For real world photography there’s not much happening on here. In fact, if anyone dares relate any of this back to the real world you get told off, talked down to or insulted. 😂😂😂😂😂😂

Only if you feel that one-third stop improvement in photon noise will make a difference in your photography. The point of the numbers is not to tell you what to do. The point of the numbers is to give you information so that you can decide what to do.

My guess is that you have far more posts on "science" threads here than anything else. You may counter with "that's all there is", to which my reply would be for you to find the start new thread button and get it started. This forum is a blank canvas. If you want more image threads, or whatever you are looking for here, simply start it. I will click on it.

Or maybe I’ll just keep posting that stuff on my own blog, or maybe join FM which seems to be massively more oriented to photography rather than a bunch of scientists having a pissing contest