Digital ISO vs Post-Exposure Correction

Question

As far as I know, a ISO setting on a DSLR camera corresponds to a mixture of both analog and digital settings set before a shot is taken and governs how much amplification the incidence luminosity is to be amplified by to obtain the final data stored in the RAW file.

On the analog side, this corresponds to various gain settings at the sensor and ADC sites to boost the effective voltage to higher/lower levels. On the digital side, this corresponds to some specific technique used to increase the value of the digitized luminosity value to a higher level before it is stored into the RAW format, all done in the digital side of things.

When composing shots, I've always taken the approach to minimize the digital ISO applied onto images and instead rely on post processing to boost exposure. To clarify, in a scenario where the correct image sensitivity setting lies between two native ISO settings I would be tempted to lock my ISO to the lower native sensitivity, taking the picture slightly underexposed, and boosting it in post (as opposed to taking it with correct exposure directly).

However, I've noticed that this procedure seems to result in lower quality images compared to in-camera digital ISO boosts. The processed shots in lightroom tend to develop a purple tint as well as seem noisier in general.

The following demonstrates two sample shots in which the exposure of the ISO 6400 one is raised by 1 EV in post via lightroom. Both shots are 100% crops of a condenser microphone's grill.

Notice the purple hue from the 6400 shot as compared to the 12800 shot. Funnily enough, the 12800 shot seems more usable despite being considered to be an "expanded" ISO for the EOS 7D (Though both shots look terrible).

Given that digital ISO boosting happens, well, digitally, why is there a disparity between the digital ISO boost applied in camera vs the one applied in post by Adobe Lightroom? I would be have assumed the reverse where the post-processed image is superior due to the increased processing power and lack of realtime preview requirements that camera processors have.

I have thought about the whole theory of exposing to the right and have tried such a theory whilst maintaining my habit of locking to the nearest native ISO but the results were not much different to the ones shown above (purple hue, apparent noise increase, etc).

Answer 1

Canon 7D ISO 12800 does not look like it is pure digital ISO. The raw histogram lacks those gaps at each second bin which are a sign of 2x multiplication. Also, Lr often applies ISO-dependent noise reduction; and there are other converter-dependent effects too.

Answer 2

That is due to the signal-to-noise ratio being higher when exposing properly than when you underexpose and raise exposure on a RAW processor like Lightroom. Setting the right ISO in-camera produces higher quality shots than raising the exposure in post-processing.

If perfect exposure can't be achieved, you either overexpose or underexpose.

I personally expose to the left (underexpose) because recovering shadows is much easier and cleaner than recovering highlights.

This should explain signal-to-noise ratios: http://www.cambridgeincolour.com/tutorials/image-noise.htm

Answer 3

Since there is only one stop difference between the two photos in your example, I would be very prone to say the largest amount of influence over the differences between the two images is the difference between the way the in camera processor handles "extended" ISO settings versus the way LightRoom increases the exposure of the ISO 6400 image. They are not the same. Sometimes, depending on the specific camera model, they cam be quite different. Just the difference in the default behavior for noise reduction of high ISO images might explain the majority of the differences between the two shots in your sample.

You would probably get much closer if you compared the in camera results to the camera manufacturer's RAW convertor. The increased processing power is only useful if the RAW file is being interpreted by an algorithm as good or better than the one used by the in camera processor.

Answer 4

You are doing it wrong. You always want to use the best ISO you can from the camera and avoid amplification in post-processing.

The reason for this is that the sensor amplifiers are analog amplifiers which will more faithfully amplify the signal than is possible after you go through the ADC. The reason for this is simple, let's imagine you have a 12-bit sensor, 4096 different values, and the scene is really dark so the tonal values are tiny, like 1, 2, 3, 4, 5, 6 etc. So imagine we have image data like this:

2, 2, 4, 2, 0, 0, 5, 1, 2

Ie, really low values, it's dark. There will be noise, so for example, maybe the 5 should have been a 4 and the 4 should have been a 3 or whatever. Now if you just amplify this 4x in photoshop you will get this:

8, 8, 16, 8, 0, 0, 20, 4, 8

It just multiplies the values. No new information. However, if you increase the ISO and get analog amplification, you might get data like this:

8, 9, 14, 6, 3, 1, 17, 3, 7

So, this data might be noisy too, but you are getting much more accurate tones and a lot of new information. The basic rule is simple:

Set the ISO to maximize the dynamic range of your sensor.

Basically you want the ISO to be as high as possible without any oversaturation of any of the pixels. You can check the histograms to see if you are getting oversaturation. Even better is to use channel-specific filters which can block excessive light and allow you to pump up the ISO even further. For example, if you have a lot of excess blue light, you can use a yellow filter to mute it, then you can increase your ISO further without over saturation.