DxOMark's measured ISOs vs. manufacturer ISOs

Started Mar 13, 2013 | Discussions thread
GordonBGood Veteran Member • Posts: 6,308
Re: ISO vs "pushing"

edhannon wrote:

Gordon, your reply got me to thinking about ISO and digital. In particular, the question of whether, when faced with a low light situation, if 'tis best to change ISO or to do like we did in the film days - expose as though we had higher ISO in the camera and push develop?

I have never been really satisfied with the testing to answer this. Concerned that too many variables changed between the two comparison images - noise reduction, ettc.

So for my question: would it make sense to measure this way:

1. Create a tone curve for at the base ISO by taking a succesion of images of a grey card at varying exposures and plotting the result. I still have my program that plots this from the origianl raw files. It also identifies the the SNR 0 point.

2. Create another tone curve at a higher ISO (say 800)

3. Compare number of stops from metered grey to SNR of 0 for both. Subtract the number of stops difference between ISO 100 and 800 (3 stops) from the base ISO one

If the higher ISO has more stops than the adjusted base ISO then increasing the ISO is best. If the adjusted base has more stops then "push" developing in Lightroom/ACR is best. If they are about the same do what ever is easier.


Ed, I think the question of the differences in image quality between underexposing a low ISO shot and "push processing" or to advance the ISO sensitivity of the camera has been fairly well answered but it depends on the camera.

Remember that "push processing" produces the the same sensitivity gain as the camera applies in camera to boost ISO sensitivity other than that the lower ISO's are generally boosted in the camera by analogue gain whereas higher ISO sensitivity gains are achieved the same as "push processing" by just a digital multiplication - the effect is the same and the reason analogue amplification is necessary is due to limitations of the data acquisition circuitry as to injecting noise between the sensor and the digitization.

For cameras that effectively do not inject noise (or use circuit means to cancel it out), digital multiplication has the exact same effect as analogue amplification as long as there are sufficient bits used in the digital capture to express all of the usable quantization levels in the signal above noise; however, for those cameras that have injected noise between the sensor and the digitization, analogue amplification works better for the first few stops in sensitivity gain as this amplification amplifies the signal without amplifying the injected noise as digital multiplication would.  Examples of this are pretty well all Canon DSLR's and are clearly seen in the DxOMark Dynamic Range (DR) charts as a nonlinear response with decreasing ISO sensitivity as they tend to reach a DR level limited by the injected noise.

The test that you are proposing essentially just repeats the testing done by DxOMark in determining these DR charts, and your results should be just about exactly the same as long as you allow for all the factors for which they allow such as that the very lowest level signals may be clipped to zero with the negative excursions of noise "blocked" if the raw data format has already applied black level compensation to offset black to level zero rather than some positive offset as is the normal output from the sensor.

So yes, you could do the testing but it is much easier to just refer to the DxOMark DR charts where the work is already done for you.

The test as you propose also has some extra variables that can cause some concern - for instance in order to test a camera with a 14 stop DR one would need to use shutter speeds for a given aperture over this same 14 stop range so that for a typical camera with a 1/4000 second fastest exposure time, one would need a four second exposure time; in order to avoid these slow exposure times below about a tenth of a second, one would need to use aperture as part of the exposure control which would add another variable.  DxOMark does this testing using single exposures taken of a type of step wedge so that only one exposure needs to be taken per ISO sensitivity.

Regards, GordonBGood

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow