7
\$\begingroup\$

Firstly, I understand the idea behind HDR photography so I'm not looking for an explanation of HDR. What I'm interested in is the large number of exposures often used to capture a scene with high dynamic range. I understand that if there are three stops more of scene dynamic range than the sensor can capture that it makes sense to take an exposure that exposes for shadows and mid-tones and lets the three brightest stops clip and take a second exposure that is adjusted to ensure those highlights don't clip. This would appear to cover the full dynamic range of the scene. However in this situation, people often take extra exposures in between the two outer brackets. For example they might expose to a mid-point and then bracket at +1, +2 and -1 and -2.

Given that in this situation the two exposures overlap by seven stops, what is the advantage (other than safety bracketing) of extra exposures in between? If processed to increase dynamic range algorithmically rather than manually, is there a technical advantage to these extra frames? Often when I see people process multiple exposures manually, they will take the most correctly exposed image and blend in another image that captures the areas that were outside of the default image (for example the sky).

So long as the full dynamic range of a scene is covered by two images, what are the advantages of capturing more images in between?

\$\endgroup\$

2 Answers 2

13
\$\begingroup\$

There are quite a few advantages. A difficult problem that often arises is blooming of bright areas into adjacent dark areas. So, the overexposed pixels that are in the bright area will leak electrons to adjacent pixels, making them get gray values that are too high. If the contrast is very high, those pixels may be in dark areas. This means that with only two exposures, you'll have to deal with blooming artifacts which shows up as a weak ghosting like effect surrounding bright areas. If you have additional intermediary exposures, the blooming artifacts can be greatly reduced.

Noise is mentioned by Chris in his answer; the more exposures you use, the better the signal to noise ratio will be as the intermediary exposures allow you to use a higher exposure for parts of the image compared to just having two exposures. Also, in the areas where more than one exposure can be used, they can be averaged which reduces the noise more.

Even if you have two images one with more noise than the other, it's still possible to reduce the noise below that of the best of the two by taking a weighted average, the optimal weight when the pictures are normalized to the same brightness and have the same absolute noise, is to take the weights proportional to the square of the exposure. This means that for unnormalized images you should give the images a weight proportional to the exposure.

Another advantage is to have more opportunities to remove changes in the scene that you either don't want to appear in the picture or which could throw off the HDR computations. Suppose that a car with bright lights appears in the scene when you take one of the exposures. Because the bright lights don't appear in the same spot in the other exposures, this will cause artifacts to appear there. If you have more than two exposures you can use another exposure to deal with the affected area.

\$\endgroup\$
6
  • 1
    \$\begingroup\$ Great answer. Thanks very much. So by this argument, even in a scene with limited dynamic range (enough for the sensor to cope with without clipping), it actually still makes sense to bracket for shadows because 1) of better SNR and 2) Possibility to reduce any noise through averaging? \$\endgroup\$ Feb 9, 2016 at 19:03
  • \$\begingroup\$ @Pedr That's right. It's best to use the expose to the right (ETTR) method that Chris mentioned. Usually I take multiple such images to average out the noise. If I take HDR images, I'll try to do take each exposure multiple times to average out the noise. Note that exposing x times longer will improve the signal to noise ratio (SNR) by a factor of x, but if you average over N images, the signal to noise ratio will improve by sqrt(N). Buy doing ETTR, you can usually expose for 3 times longer. If you take 6 such images then the net gain in the SNR by sqrt(6)*3 = 7.3 \$\endgroup\$ Feb 9, 2016 at 19:41
  • \$\begingroup\$ I do expose to the right in a lot of situations, though i've never done so when shooting an HDR sequence, but it makes a lot of sense. The only thing with ETTR is whether you are introducing more noise by getting there (via increased ISO), but I'm guessing you would say that any extra exposure should be done through aperture and shutter speed rather than ISO? \$\endgroup\$ Feb 9, 2016 at 19:47
  • \$\begingroup\$ Yes, it's better to expose for longer instead of increasing ISO. Also, I always shoot at a fixed aperture, I choose the one for which my lens is the sharpest. A problem with using different apertures is that the HDR stack will have to be aligned, at the edges of the images you'll get problems with alignment, also there will be unsharpness that is not the same in the different images, which then leads to ghosting. \$\endgroup\$ Feb 9, 2016 at 19:55
  • \$\begingroup\$ Sorry. Wasn't clear. Not suggesting changing aperture during stack, just pointing out that if ETTR is only achieved by increasing ISO then it introduces its own noise issues, so in order to achieve ETTR it only really makes sense to first open up your aperture (as you say the sweet spot will depend on the lens), then increase shutter speed as needed. I have yet to see a comprehensive study into the benefits of ETTR vs the negatives of increased ISO - in situations where a long exposure isn't an option whether it is better to increase ISO to allow for ETTR or keep it low and forego it. \$\endgroup\$ Feb 9, 2016 at 20:00
5
\$\begingroup\$

More images will give you better signal-to-noise ratio (SNR), since you will have a good SNR for every absolute brightness value if you do more intermediate steps. The brightest captured parts of the image will have the best SNR and, thus, “details”, it's the same reason for “expose to the right” (see https://en.wikipedia.org/wiki/Exposing_to_the_right). If you do 1-EV-steps, the algorithm can take the best 1 EV range of every picture but the darkest one. Therefore it may be beneficial to add some steps at the lower end if possible.

\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.