Is there a situation in which taking exposure bracketed photos at only the widest separation (±3EV) doesn't work?

Question

I sometimes shoot exposure-bracketed photos on my Sony NEX-5R, for exposure fusion in Photomatix.

I'm thinking of taking just two photos, at -3EV and +3EV, which are the maximum values of exposure compensation the camera allows. I think that there's no need to use a smaller separation, like 2 EV, and no need to take more than two photos. My reasoning is as follows:

This camera has a dynamic range of 13EV, which is much greater than the 6 EV separation between the +3EV and the -3EV photos. This means that there's 7 EV (13 - 6) overlap between the two images: The brightest 7 EV of pixels in the image captured at -3EV will be the same as the dimmest 7 EV of pixels in the image captured at +3EV. This is essentially redundant information. Taking more than two photos, or taking photos separated by < 3 EV means greater overlap, and more redundant information captured, which doesn't help.

Is this reasoning correct?

Footnotes:

1) In low-light situations, a photo at +3 EV may take a long time to capture, like 30 seconds, which you may not want if a photo taken at +2 EV will do the job.

2) I'm aware that DxoMark calculates the dynamic range at base ISO. Assuming that the dynamic range is only 7EV rather than the 14EV DxoMark measures, that's still greater than the 6EV separation between a photo taken at -3EV and a photo taken at +3EV.

Answer 1

You are incorrect in the assumption that redundancy is a bad thing. A good HDR tool will be able to average the results of many images to reduce noise. Having more images also reduces the chance that there is movement in one of the images that will impact the result.

A few points, DXO also measures dynamic range at each ISO setting. The NEX 5R has 13.1 EVs at ISO100, which doesn't fall to 7 EV until ISO 12800!

Finally I would question the benefits of HDR at all with a Sony Exmor sensor. At ISO100 you are already getting 13 stops, if you do +-3EV that's 19 stops, which is almost certainly greater than the lens is capable of delivering. All lenses suffer from veiling flare to a certain degree which will prevent you digging too far into the shadows.

Answer 2

The reason it makes sense to use smaller brackets is that you get better detail when dealing with a scene with less dynamic range. Any part of the range of the image that isn't used is wasted. You want to be sure your images are hitting the black and white points for maximum benefit.

If an image only covers a 17EV range, then your best result will come from covering that 17EV range rather than covering a 19EV range and having less overlap between the images. You will lose an two entire EVs of extra detail in the middle since you spread them out further and have less overlap between the images.

Details that were weak in the highlights of one could be filled out by having strong detail in the midrange of the response with less overlap but would otherwise be crushed to black or white with a wider split.

Answer 3

I realized after posting the question that I could do an experiment to prove or disprove the two claims in the question:

1. That you can always take a burst separated by 3EV rather than smaller values.

2. That it's enough to have two photos, rather than three, as long as they cover the dynamic range of the scene.

Regarding the First Claim:

I took a burst of three photos separated by 2EV, that is, -2EV, 0EV and +2EV, and fused them:

I then threw away the -2 and +2EV photos, and re-created them synthetically from the 0EV photo by pushing Lightroom's exposure slider up and down. I then fused the three photos:

Notice that there's a negligible difference. The conclusion is that there's no need to take a burst of photos separated by 2EV. You can instead take a single photo, and recreate the -2 and +2EV versions in post if you need them for exposure fusion. If you're taking an exposure-bracketed burst, you might as well take them 3EV apart.

---

Regarding the Second Claim:

I took three photos separated by 3EV, and fused them:

I then fused the +3EV and -3EV photos, ignoring the 0EV photo:

Fusing three photos produces a better result than fusing only two:

• The former has a better contrast between the sky to the left and the wall below the air-conditioner, while the latter looks as if there's a haze permeating the whole photo, bringing everything down to the same color.

• In the former, the railing on the left looks black, while the latter looks washed out. This goes for the two corners of the balconies you see, one to the left of the air-conditioner, and the other below the first one (above the handrail).

You can see this by opening both images in two tabs and switching between repeatedly. This proves that you're better off fusing three photos than you're fusing only two.