Hi Feppe,
I distrust common "wisdom," especially in photography. Do you have some research backing up the claim that 4 stops apart is not optimal for exposure blending? Guillermo Luijk has an excellent article on the topic which is my basis for using 4 stops.
4 stops apart is quite large. Let me explain:
Exposure compensation value is calculated as the base-2 logarithm of the ratio of the exposures of two photos. In other words, the pixel values in the first photo are 16 times lower than the ones found in the second photo (the inverse of base-2 logarithm is 2 to the power of the number of stops, here: 2^4 = 16).
Camera censors have a "sweet spot". Below a certain low value, data is buried in noise. After a certain high value, the censor response to incoming light becomes non-linear.
The goal of HDR is to use the range of values between these two extremes for each of the photos you are assembling. The algorithm attempts to use only the best part of each photo in terms of exposure.
Here is an example:
- Photo 1: regular exposure
- Photo 2: under exposed of 4 stops (-4EV)
Assuming 8-bit images (the reasoning is the same with 12 or 14 bits but easier to follow).
- Total range of pixel values is: [0 255]
- Useful range is, say: [50 200] (under 50 there is too much noise, after 200, signal becomes non-linear)
For both photos, the algorithm will use the range [50 200].
- Photo 1: [50 200]
- Photo 2: [50 200]
Now, the Photo 2 is 16 times darker than Photo 1 (4 stops). Because of that, pixel values of Photo 2 must be multiplied by 16 before assembling with Photo 1:
- Photo 1: [50 200]
- Photo 2: [16*50 16*200]
That is:
- Photo 1: [50 200]
- Photo 2: [800 3200]
As you can see, the resulting HDR image has values ranging from 50 to 3200. Highlights in Photo 1 are fully recovered. There is anyway a serious issue: where are the values from 200 to 800? Answer: nowhere! There is a large gap between the range [50 200] and [800 3200]!
At this point, you should start guessing why the "2 stops rule" makes sense in HDR photography!
With 2 stops, the exposure ratio is 4 (2 to the power of 2). Let's take the previous example:
- Photo 1: regular exposure
- Photo 2: under exposed of 2 stops (-2EV)
Ranges after blending:
- Photo 1: [50 200]
- Photo 2: [4*50 4*200] = [200 800]
The HDR image has now a range of [50 800] and this time there is not gap from [50 200] to [200 800]. To get the same dynamic range than the first example, you could take three photos 2 stops apart:
- Photo 1: [50 200]
- Photo 2: [4*50 4*200] = [200 800]
- Photo 3: [16*50 16*200] = [800 3200]
You end up with a continuous range of [50 3200] without any gaps.
The previous discussion is pure mathematical reasoning (and I have simplified quite a lot what is really happening in HDR / Exposure Fusion). Does it really matter to practical HDR photography?
Most of the time, the answer is yes, but not always!
The light in most scenes increases smoothly from the darkest shadows to the brightest highlights. Introducing a gap in the tonal range is not a good idea. You should then use the 2 stops rule. I even recommend setting the photos 1 stop apart for best results (in that case, there is a tonal overlap between each photo that is beneficial).
In the case of night photography, there are mainly two ranges: very deep shadows and highlights (such as street lights). In this special case, there is a natural gap in light intensities in the real world scene. Using 4 stops can make sense in this situation. I would anyway recommend that you use 2 stops or less for other reasons (such as additional image stacking that will reduce noise but that is another long story!)
Regards,
Antoine Clappier