Today I repeated the whole series of noise-vs ISO setting tests for the Leica M9.
First, a check for self-heating with a series of repeated 1/4 second exposures of blackness at maximum ISO. Here’s the average value of the noise floor over the series:
And here’s the standard deviation (the noise of the noise) over the series:
Essentially no self heating. The Leica can’t make continuous exposures very fast, which might explain some of that.
Then a test of the noise floor versus the ISO setting at 1/30 of a second, which is the exposure time I’ll use for all the following tests. Sixteen exposures for each data point. Ninety percent of the image area in the sample. First, the average value of the noise floor:
Second, the standard deviation of the noise floor:
Next, the noise in at 200×200 pixel centered patch at 1/30 of a second, with the lens opening and the target brightness set to give Zone VI counts — about 4000. The lighter lines are plus and minus two standard deviations. These graphs include noise floor corrections as described previously:
The above graph with half a stop per stop subtracted out:
It is clear that camera ISO settings of up to 640 help a little over just cranking up the Exposure dial in Lightroom, and that ISO settings above 640 actually do harm.
Let’s look at Zone III — about a count of 500:
Alf Sollund says
Thanks a lot for sharing. So in essence one should treat the M9 as ISOless from 640 (or 500) unless you need shorter exposure time?
I’m not sure exactly what you mean by “treat the M8 as ISOless”, so I”ll try saying it another way. If you can’t get the histogram on the M9 to the right with f-stops and shutter speeds that are acceptable to you, you can move the histogram to the right by setting the ISO higher, but stop at 640. If the histogram is still not as far right as you’d like, don’t worry about it. You can fix it in ACR or Lightroom with better results than you’d get if you further increased the ISO setting.
Does that make sense?
I saw your post on the LCamera Forum, and now I understand what you mean. Yes, you are exactly right. I could do a series of tests at one-third stop ISO setting spacing to get at your implied question about ISO 500, and I will if there’s sufficient interest. I am somewhat reluctant because the small buffer size and slow write performance on the M9 makes this quite time-consuming, especially with 16 shots per data point. Right now I’m working on the Sony RX-1 and the NEX-7.
Duane Pandorf says
Makes perfect sense to me.
Thanks for the test, Jim. I’ve been shooting my M9 pegged at ISO 160 in all light, and then just boosting in LR4, so it’s good to know that the in-camera ISO does actually improve things up to ISO 640.
You’re certainly welcome. You’ve been working that way for a while, so I’m sure it’s clear to you, but for others who wander onto this thread, I want to be clear:
You’ll get the best results at ISO 160 with ETTR. If doepth of field, subject or camera motion, or other things keep you from doing that, you’ll get slightly better results in the mids and higher tones by turning up the ISO to keep the histogram to the right than you will pushing equivalently in LR or ACR, and you’ll get essentially the same in the shadows. That’s only true up to ISO 640. After that, let the histogram go to the left and fix it in your raw developer program.
Yeah, I used to go up to ISO 640, but, in my unscientific tests, I had a difficult time distinguishing between the ISO 640 and ISO 160 pushed results, so I’ve been keeping things at ISO 160 to make things easier and avoid any risk of blowing highlights. I’ll have to start using ISO 640 more.
One other thing to consider, in terms of ETTR at ISO 160, is that sometimes exposing midtones too far to the right can affect color. Have you had a chance to test this at all? The Chromasoft blog talks about this from time to time.
If you look at the highlight SNR improvements up to ISO 640, you’ll see that you gain less than half a stop improvement by increasing the ISO setting over pushing in post. You don’t gain anything in the shadows, which is where the noise is most visible. So, if you’re happy with your current method, I don’t see a lot of need to change it. One problem with pushing in LR or ACR is that you can only get a five stop push. So, if things are really dark, you could run the ISO up to 640 so that you have the latitude to do the rest in LR.
Good points. Of course, the +5 EV limit in LR4 gives me a nice end point, because things get pretty noisy when you get past the equivalent of ISO “5000” with this camera. It has kind of been my artificial limiter. 🙂
WRT ETTR color shifts, it is true that ETTR makes you do more work in post than you normally would. LR and ACR aren’t really set up to deal with ETTR images. IT’s amazing how well the new Exposure control works, considering that we’re using it in ways probably not intended by the developers.
Yeah, I don’t shoot commercially anymore, so, for my own work, absolutely accurate color isn’t essential, and I tend to play around with it, anyways, so potential color issues from ETTR probably wouldn’t bother me much. Thanks!