The Truth about 1/3 ISO stops... and how to apply with E-5

[GWJ]

There was a very interesting discussion about 1/3 ISO stops in the Olympus Flickr group. Basically the heart of the discussion, is that 1/3 stops are software driven, and full stops are part of the hardware. I own a e-5, and now that I have disabled 1/3 stops I have lost 2500, which I felt had some usefulness above 1600. I notice my e-5 says 3200 and above is an extension.. I feel iso 2500 retains a bit more detail than iso 3200, but I wonder if I am introducing more noise than if I stick to full stops in iso. And I also read that everything below iso 200 is an extension, and dpreview reviews of the e-5 suggests that owners of the the e-5 should not go beneath iso 200. I have used iso 100 and I find it much cleaner in noise that iso 200, dpreview suggests that the cost of this is reduced DR.

What do others think of this discussion.. and what are other e-5 owners using

Here is the link

http://www.flickr.com/groups/olympusesystem/discuss/72157626252480834/

 

[Gregm61]

There's always been a ceiling in terms of high ISO settings. The early models made you manually turn on the extension, otherwise, the E1 stopped at ISO 800 and the E300 stopped at ISO 400. The new models just enable it automatically.

I heard about this story but have seen little to make me change from using the intermediate settings. ISO 2000 is my preferred manual setting for shooting hockey at the local NHL arena. That won't be changing and I'm looking forward to taking the E5 to a few night major league baseball games as well and will use whatever setting it takes, intermediate setting or not.

 

[jkrumm]

I would not worry about it. All the iso levels are "software driven" to some extent. Use what works and looks good to you. Disabling the 1/3 steps just gives you fewer options and makes auto-iso less useful.

--
John Krumm
Juneau, AK

 

[LinusP]

I can't speak to the E-5, but I switched from 1-EV ISO steps to 1/3-EV ISO steps on my E-620 several months back. In good light, I usually shoot in aperture priority, and use ISO 100, 200, and 400 almost exclusively (I've trained myself to do three clicks of the wheel almost as fast as I used to do one). As the light levels drop, I switch into manual exposure and switch to Auto-ISO, and that's really helped improve my percentage of keepers. It seems I'm only capable of managing two sides of the exposure triangle at once; having Auto-ISO is god-send for me.

 

[sanjaykool]

1/3-EV ISO steps can give you 0 to 2/3 EV advantage.

This depends on how close is ISO to full 1 -EV stop.

However unlike film digital sensor works on amplification of signal in continuous mode.
That way every thing is hardware driven, whereas computation is software driven.
That is why more noise at higher ISO for every camera sensor.

Why to disable it ? At least that is my understanding.

 

[acsmith]

You might want to take a look at my e620 test shots and analysis over at Steve's: http://forums.steves-digicams.com/olympus-dslr/185525-1-3-iso-noise-examples.html .

I really expect that the e5 will show the same pattern of noise increases as the e620 as the main difference is the lack of anti-aliasing filter. That pattern is that at any full stop multiple of the base ISO has less noise than the 1/3d stop less than the full stop multiple. That is 2500 has more noise than 3200, 1250 more than 1600, 640 more than 800 and so on. On the other hand the 2/3d stop under a full stop base multiple has less noise than either the 1/3d stop above it or the base multiple above it. That is 2000 has less noise than either 2500 or 3200 (but more than 1600 of course), 1000 less than 1250 or 1600, 500 less than 640 or 800 and so on. Unfortunately there is no way to apply that automatically, that is base multiple and the 1/3d stop above.

Gregg's selection of 2000 for his NHL shots has a sound basis and you should go over to Steve's and take a look at his hocky work.

Finally, visual analysis suggests that the non-linear noise increase at the 1/3d stops is primarily chroma noise so if your PP allows seperate adjustments for luminance and chroma noise you might want to apply more filtering to chroma noise if you're stuck with 1/3d stops.

A. C.

--
I've taken a vow of poverty. To annoy me send money.

 

[Claus A]

Hmm,

I recall a 'test' by Robert Schroeder (not with the E5, however) on the 1/3 ISO steps.

He photographed the identical scene with different ISO steps and compared the file sizes (increasing ISO introduces noise, thus the file will be larger). The 1/3 steps were 'real' steps in that test.

Can't say how it's with the E5, though.

Regards,

Claus.

--

... when the photograph annihilates itself as medium to be no longer a sign but the thing itself...

 

[acsmith]

Hmm,

I recall a 'test' by Robert Schroeder (not with the E5, however) on the 1/3 ISO steps.

He photographed the identical scene with different ISO steps and compared the file sizes (increasing ISO introduces noise, thus the file will be larger). The 1/3 steps were 'real' steps in that test.
Claus.

Oh the files sizes increase alright but you have to compare 160 vs. 200, 640 vs 800, 1250 vs 1600 and 2500 vs 3200 to really see what's going on. The real issue is how is the ISO being increased? The whole stop increases seem to be handled by increasing the analog gain on the sensor output while the intermediate 1/3d stops are being handled by digital mathmatics which increases noise faster than would be predicted by simple analog gain changes.

My test images were the inside of the lens cap and since I'm not DPR or DXO lab my only quantification method was to use the file size size and that's the results posted over at Steve's. If anyone has software capable of doing a better analysis or seperating luminance noise from chroma noise they can download the my images from Steve's or if they have an FTP server I'll upload the full raw or jpg files to their server.

A. C.

--
I've taken a vow of poverty. To annoy me send money.

 

[Velocity of Sound]

I missed your post here initially... I was considering bringing this topic up on DPreview once the data was more final.

As a bit of background, I'm using an image analyzer tool from the NIH called ImageJ (it's free). In research, we have used this to analyze the intensity of scanned western blots, as a means of measuring relative protein concentration (it's relative unless you have loaded controls that you can correlate measurements to). I figured that the program could be used as a method of quantifying the amount of ISO noise. However, it does not characterize the noise.

You can check the Flickr thread if you're interested in the testing methodology. What you may be interested in are the results:



Three shots were taken per ISO value; the bars (and the numbers beneath them) represent the average. Error bars indicate the standard deviation between the three shots.

The take-home message to me was that the E-3 (and probably cameras before it) experiences worse performance at 1/3 ISO steps below ISO 800. However, above ISO 800 the 1/3 steps seem to provide a benefit. With the E-P1 it's somewhat similar, but even with the increase in noise from 1/3 steps below ISO 800, the amount of noise that we're talking about is pretty minor. (The E-P1's noise performance also seems to be much more consistent than the E-3's.) I would hazard a guess that all cameras produced after the E-P1 (including the E-5) mirror this trend, but without analyzing the RAW files I can't say for certain. The big question is which trend the E-30 and E-620 follow, as they supposedly use the same sensor as the E-P1, but some initial evidence suggests that they follow the E-3 more closely.

Someone above questioned why the noise should be worse at 1/3 steps. The reasoning is like this: the normal ISO steps are implemented in the hardware, and the setting adjusts the sensor's actual sensitivity. The 1/3 steps result in greater noise because they supposedly are done in software. In other words, a shot at ISO 640 is actually done at ISO 400 - which would be underexposed - but the camera then increases the exposure in software. If you have ever tried this yourself, you have probably seen that you bring out noise by pushing the exposure in such a manner.

Nobody aside from the engineers in the companies knows for sure if that's what is being done, of course - it's a theory.

If anyone with another camera model is willing to take at least three shots per ISO step and send the RAWs to me for analysis, please get in touch with me. I will give you a more specific set of instructions for how to take the photos in a manner that matches how I did it, so as to remove as many variables as possible.