Looking over a few RAW files from various Canon dSLRs, I noticed that some, not only appear noisier, but also reveal slightly different colour shading when using in-between ISOs (125, 160, 250 320, 500, 640, 1000) compared to native ISOs (100, 200 400, 800 and 1600).
So I decided to do a brief bit of internet research and read a few articles that suggest that Canon (and others?) actually employ push and pull techniques post A/D conversion to derive these in-between ISOs.
In other words, it’s like shooting at ISO 400 and then increasing the exposure by 1/3 stop after the event, to yield ISO 500, or shooting at ISO 800 and reducing the exposure by a 1/3 stop to yield ISO 640.
So using three dSLRS (5D, 1DMkIIN and 1DMkIII), I decided to run a series of black noise tests to validate this phenomena.
Before I get into the detail, here’s the some background information on the methods employed.
Each dSLR’s sensor was checked and cleaned and the body cap was left in place. The eye piece was covered and the shutter was set to 1/8000s.
All noise reduction functions were disabled and so was highlight tone priority (where applicable). ISO expansion was enabled on all cameras.
I shot two frames (averaging) for each ISO at 1/3 stop increments. In order to minimise any internal effect that may arise whilst writing out the image to the CF card I maintained an interval of 4s between shots.
In turn, each RAW was loaded into IRIS and centred on my NEC 26″ LCD at 100% zoom. Using a ~1920 x 1600 pixel selection, I then sampled and noted the Sigma value.
For 12 bit files (5D and 1DMkIIN) the Sigma values were multiplied by 16 to yield 16 bit results, while the 1DMkIII values were multiplied by 4 to yield 16 bit results.
Findings
For the 1DMkIII and 5D it was interesting to note that ISO 50 is really no more than a ND filter, I got identical noise readings for each camera at their respective ISO 100.
For the 1DMkIII, I am now reluctant to use ISO 125, 160, 250, 320 and 640, but would prefer to bump to the next native ISO e.g. 200, 400 and 800, as the results show significantly less noise.
ISO 500 on the 1DMkIII does seem to be a bit better than ISO 800. So the extra 1/3rd of stop over ISO 400 could be seen as a marginal gain (if you need a slightly faster shutter speed).
For the Canon 1DMkIIN, ISO / noise variations seem to be subtlety different.
For starters, ISO 50 shows a marginal improvement over ISO 100. Also, the steeper noise increment starts at ISO 400 as opposed to ISO 800 on the MkIII. While the 1/3 stop pushed ISOs (125 and 250) reveal marginally less noise than the next full stop.
So the usable ISOs on the 1DMkIIN are 50, 100, 125, 200, 250, 400….etc
Comparing the 1DMkIIN v 1DMkIII in terms of noise, the 1DMkIII has ~1.6 stop advantage, with ISO 800 appearing almost as clean as ISO 250 on the MkIIN. This advantage narrows to <1 stop at higher ISOs.
For the 5D, usable ISOs are again mainly native e.g 100, 200, 400, 800….etc
In comparison to the 1DMkIIN, the native ISOs of the 5D have >2/3 stop noise disadvantage up to ISO400. Although this loss becomes a gain of nearly 1 stop at the higher ISOs.
This may appear strange but the 5D’s read noise level seems flatter across the non expanded ISO range. Although my 5D does appear to be noisier than both 1 series at the lower ISOs.
In comparison to the 1DMkIII, the 5D has ~1 stop noise disadvantage up to ISO400 and this reduces to ~1/3 stop disadvantage at the higher ISO (3200).
Conclusions
For general lighting conditions, In-between ISOs perform poorly up to ISO 800 (ISO 400 on the 1DMkIIN), there seems to be no real advantage in using them and therefore (in my opinion) it’s better to switch off the 1/3rd stop ISO increment in the custom function and stick to native ISOs.
However if your photography involves low light work, then working above ISO 800 (ISO 400 on the 1DMkIIN), then the noise increment found at 1/3rd ISO intervals yields better results.
