HDR – How Many Exposures are Enough?

How many exposures are enough?

When it comes to shooting HDRs, one of the biggest questions asked is, how many exposures should I take and how far apart should they be spaced. Everyone has their opinions and I’ve seen people go everywhere from 19 exposures down to…well 1. With spacing all over the place from the uber-anal 1/3 stop to people just spacing them randomly.

I’ve explored this before in blog posts of the past but I thought I would take a look at it again in a slightly different way and I thought I would take some images from this past weekends shoot at the surreal Salton Sea and put them to  as scientific a test as I could.

Now if you have a method that works for you that’s great. I’m not here to change your mind or your workflow. But this is for people that don’t know or are confused by all the options. It’s taken mostly from my perspective: I don’t like to do things that waste my time or energy for little or no gain, but I will go the extra mile for something that pleases me and obtains my goal.

Here is a little of the methods of my test. All images were shot in Raw on my Canon 5D with a 17-40mm L lens at 17mm. Both scenes’ dynamic range was measured by the camera meter and the maximum number of images at 1EV spacing to cover that range were taken.

All images were imported into Lightroom and no adjustments were made to the images. The images were exported to Nik HDR Efex Pro 2 in varying Numbers and EV spacings making final HDRs for each. No alignment, deghosting or CA adjustments were used since they can vary the look of an image. The same Preset, that I custom made for each test Scene, was applied to all images of that Scene to keep the look constant.

The images were then taken into Photoshop and the same Levels adjustment layer was applied to all images to make an image look as close as possible to a totally finished image with the exception that no Noise Reduction or Sharpening was applied to any image so we could see the full range of noise from different scenarios…whew..Ok…let’s go

Real world testing

OK, I have to say there were some thing I was hoping would happen, but you can’t do a scientific test if you WANT a desired outcome. You have to look at it honestly and open mindedly. I have to say I was surprised by some things and then other things just make pure sense

Here is our first scene. This scene is very high dynamic range; using the 32 bit Histogram in Photomatix Pro the captured scene has a contrast ratio of 149,255:1 which is incredibly high, pretty much beyond that of the human eye in one look. It’s a tough scene also to shoot. It’s directly into the sun and then the subject is highly backlit which is not a great scenario to shoot under since backlit subjects don’t have the best light on them, it’s very flat and desaturated and doesn’t always make for the best images. But it is tough and that’s what I wanted. The sun in any scene always increases the dynamic range of a scene especially when it is still far above the horizon line.

I measured the dynamic range of the scene and knew I needed with my aperture at f/22 and ISO 200 to use shutter speeds from 1/4000 to 1/15. I would cover that in 1EV spacing giving me 9 Images (I also shot a 10th at 1/8 but did not use it in this experiment since it yielded very little information [blowout}

Here is what those exposures looked like

Here are the Number of exposures and Stops I tried

* 9 exposures 1 Ev (stop) intervals. This represents the entire DR of the scene  9 stops + Camera Dynamic Range

* 5 Exposures 2Ev intervals. This again represents the entire scene of 9 stops but  the interval  between exposures is wider

* 7 Exposures 1Ev interval. This drops the brightest and darkest exposures so it represents 7 stops but still 1 stop between exposures

* 5 Exposures 1Ev interval. This drops the 2 brightest exposures and the 2 darkest for a 4 stop range, with 1 stop in between

* 3 Exposures 2EV interval, represents again the same 4 stop range but in 2 stop intervals

I processed all 5 images and the results actually surprised me. I really thought I would easily be able to distinguish between the 9/1 image and the 3/2 image and it simply, at normal viewing size and distance ,was not that easy to see.

I really thought, especially the area around the sun, I would see a huge difference, I thought I would see ore of a moiré pattern to the sky and I thought noise would just be out of control on the 3/2 image. It just wasn’t so.

Here are the 5 images and I’m going to have a little fun. I will letter the images and you try to put them in order from the 9/1 image to the 3/2 image. My little blind test. I realize you will only have a 900 pixel image to work with. But that also gives a real world viewing example. POst what order you think they should be in comments and I’ll say in comments later what the order actually is.

So why do I think this is? Shouldn’t there have been a clear winner? Here’s what I think; we tend to think of these exposures as slices 1 or 2 EV wide slices, when in fact they really aren’t. They are the full dynamic range of our cameras, each exposure represent that full dynamic range until we get to the end exposures where the dynamic range is limited by noise at the low end and saturation at the top end.  We are merely varying the zero point of the exposure and that DR.

Each of the exposures represent the full dynamic range of our cameras that vary from about 7 Stops up to almost 12 stops for the very best cameras. The DR of the camera also varies with ISO

So when we have an image like the 3/2ev image it does not just represent 4 stops, it represents about 13-16 stops when you add in the 9 stops my Canon 5D is capable of plus the moving zero point of the 3 exposures. In my thinking, this is why we don’t see very different images viewed like this.

Does this mean there were NO differences? Absolutely not, but it really took some pixel peeping at 100 and 200% to really pinpoint and see the differences.

I’ve made some 100% crops of an area of the images and you will be able to see full size if you click on the image and open it in a new window or download the image to look at in your editing software.

Most of the difference you see will have to do with noise and tonality and the results are not really that surprising

Here’s my analysis

* 9/1 image. Best image off them all, very little noise good gradation of tone and color throughout. This image should be the best one

* 5/2 image. Really close to the 9/1 image. no greater noise which is expected since the image still covers the same DR. Tonality may be the slightest bit less but I don’t know if I’m just seeing it because I now what image it is.

* 7/1 Image. Tonality is still good because of the close exposures give good gradation. However the noise level picks up which should be expected since we are minus two exposures so the software has to work a little harder to map those tones bringing some shadows up and increasing noise

* 5/1 now things really become easy to see at 100% crops even if we can’t on the full size image. Loss of tone and the noise has gone up once again because now we are missing 4 images of information

* 3/2 about the same as 5/1 noise wise but with even more loss of tone. Which should be as expected.

Even with these differences, the truth is all of the images are still very usable and we wouldn’t have e noticed easily some of these things if we didn’t have the other examples to compare too.

But these results were with an image that really had a high dynamic range scene to capture and at those times you may need to throw everything you have up against it.

What about a scene that is more typical of what I see a lot of people shoot as HDRs.

 Test  2

The next image has a Dynamic Range measured in Photomatix 32 Bit histogram of 829:1. Technically not a High Dynamic Range scene. But it has some shadow areas and like I said many people…even myself, might shoot this as an HDR. Do exposures and spacing matter on this image?

I did the same type test, the scene measure lower but knowing I was doing this test I went ahead and shot 7 exposure 1EV anyway

f/22 ISO 200 Shutter speeds from 1/400 to 1/5

The images I made were

* 7/1ev

* 5/1ev

* 3/2ev

I also took the 0 image and tone- mapped that image because honestly, this scene is not beyond the DR of my camera so I wanted to see how that image would compare.

Here are the results below in the order from above

Honestly there isn’t any visible difference between the 3 HDRs and the Single image just misses in tonality a bit.

But we should look closer once again just to see where the differences may be

While I still might crown the 7/1 image king, there really isn’t that much to talk about. All the other HDR 100% crops look pretty much the same. You can see the tonality loss in the single image and it does have more noise but that to be expected. I also included a straight out of the camera single image for comparison and the tonality really isn’t there, especially in the shadows (again expected) but that image did have the lowest noise…once again expected.

 Conclusions

The first thing to conclude, is that this test does NOT say you should always shoot 9 exposures, what it does say is that for the best image quality you should shoot the dynamic range of the scene, whether that takes 9 exposures or 1

What it also says is that; even if you don’t, you still will get a very usable image

Covering the DR of the Scene does matter, the spacing not quite as much. (9/1 and 5/2 covered the same DR) The most important part is capturing the entire dynamic range of the scene and ONLY that. There is no need to exceed it. Those images actually hinder an image, both in quality and also alignment problems. Did I really prove that in this test? No, but I have done it in others and those exposures are unnecessary. So shoot up to the Dynamic Range but you don’t need to shoot MORE than that

We also have seen that how dynamic that scene is also plays a role in how anal we have to be about capturing every last point of Dynamic range.

But we really have to take other things into consideration and for some of you these considerations won’t matter and you continue to do 19 exposures 1/3 which is fine, it’s your time.

If we take into consideration, things like:

* How will this image be used, huge print or just web postings

* Are we using a Tripod? Firing off  9 shots without one is tricky to say the least

* Storage issues, I know my cards fill up real fast with a day of 7 & 9 exposure shooting

* Processing time, it take a lot longer to merge and align 9 images than 3

* Alignment issues themselves. With poor shooting/tripod habits and 9 images you may have a loss of detail because the software just can not get everything right. It may have an easier time aligning 3

So while under the microscope there is real difference, taking into consideration everything that goes into it, you now may be able to make a better decision how you want to shoot. I know I’ll probably rethink using 1EV and switch to 2EV on difficult scenes. I already use 3/2Ev very frequently for hand holding and scenes that are not demanding and I think the test shows for that, it’s a good choice

Hope that helps

PT

Oh for those of you guessing the order for the first part, they were just as presented

A: 9/1
B :5/2
C: 7/1
D:  5/1
E: 3/2

I know no one would tell you to guess the order and then keep them in the order…but I wanted to really make it about SEEING a difference