Cheating at HDR

Over the past few months, I have shared a few examples of my High Dynamic Range (HDR) images. And I’ve gotten a fair number of questions on the technique. Now that Apple has decided to bring HDR photography to the masses via iOS4.1, I thought it was about time to post something on how I create my HDR images. I cheat. More on that later, but first…some recent HDR images:

Blue Mountains, Australia

U.S. Open 2010

Garden of the Gods, Colorado Springs

HDR photography is a great tool and can produce a wide spectrum of styles and results. In a nutshell, HDR photography involves combining multiple images of the same scene to make up for a fundamental deficiency of digital optics: cameras are nowhere as sophisticated as human vision. And when it comes to dynamic range, its no contest. That said, it does take a bit more time and patience to capture stills for making HDR images…and lots more post-processing. But in the right situation, the results can be absolutely magical.

What is Dynamic Range?

In optics, dynamic range is defined as the ratio of the maximum light signal to the minimum light signal that can be detected. In other words, the lightest tones to the darkest tones that can be seen. Humans can see a wider range of lights to darks than a camera. Much wider. That’s why our eyes can make out the facial details of a person standing in front of a bright window AND make out the leaves on the trees outside. But we’ve all experienced what happens when you ask a camera to take that picture: you’re forced to choose between properly exposed trees with a silhouette of the person OR the person properly exposed with the window completely whitewashed. How can we get both? Enter HDR photography.

Creating HDR Images

This is the focus of this post. The process, at a high level, involves capturing the right stills, then combining the images using some HDR software, and making final adjustments. I am focusing on the first step in that process and mentioning the software when it comes into play.

The Long, Hard Road (this is not the cheating part): HDR photography begins with producing a series of images that capture the entire dynamic range (lightest lights to darkest darks) of the scene. Typically 3 or more images are taken. The traditional approach involves setting the camera up on a tripod, measuring the dynamic range of the scene by exposure metering, and taking shots in aperture priority or manual mode. As I mentioned, capturing the full dynamic range requires multiple shots. To accomplish this, after each shot the shutter speed is changed…for instance, I often change by 1 stop by adjusting the shutter speed by a factor of 2 (from 1/200s to 1/100s, then to 1/50s, etc). Shooting in manual focus is also common practice (or setting the focus automatically and then switching to manual focus). Why so many rules? Because you’re taking multiple shots of the same thing, varying only the exposure from shot to shot. So making sure that nothing else changes is ideal:

  • The tripod prevents the camera from moving between images. That way you dont have to worry about alignment later.
  • Measuring dynamic range ensures that you’ll know how many shots to take to get the detail in all the lights and in all the darks.
  • Shooting in aperture priority or manual mode guarantees that your depth of field won’t fluctuate from shot to shot.
  • Manual focus helps make sure that the individual images blend as well as possible and that the focal point remains constant. Of course, you only want to focus before the first shot of the series, not between shots.
  • The software that’s used to process HDR photos (more on that later..but not too much more) likes evenly spaced exposures.

That is a lot to remember. And its even more to DO! To be honest, as a traveling photographer, this approach basically lost me at “tripod.” I don’t always have mine on me, and I certainly don’t want it to become an ever-present accessory. So…my first tip: keep your eyes open for worthy tripod substitutes. There are lots of potential tripods in plain sight. Most aren’t ideal, but I’ve been happy with the tradeoff in almost every case.

Below is an example. On a 10 degree night in January, I passed by Lincoln Center in New York City. It looked like an ideal scene to try out some HDR: bright white tones and really dark tones. I didn’t have a tripod, though. All I could find was a trashcan. So, I went inside and got permission from the lobby security guard to roll a 50 pound garbage can to a place that I liked and started shooting. Here are the individual images (in most of them, you can spot the top of the trashcan at the bottom of the frame):

I took these shots in aperture priority at f/8. I started with a shutter speed of 1/30s for the first shot and slowed it down by 1 stop increments, all the way to 4 seconds (1/15s…1/8s…1/4s…1/2s…1s…2s…4s). How did I come up with these numbers? After setting the aperture, I used my cameras exposure meter to figure out the shutter speed to expose the center of the foam in the fountain (the lightest area of the scene). That was around 1/30s. Then I metered on the gray stone on the ground (the darkest area) and figured out that about 4s would give me some detail in that area. So now I knew how many shots I would need. With shutter speeds like 4 seconds, you can see how I could never have done this without a tripod…or a trashcan. Now what?

Creating an HDR Composite Image

Here is the part that I said I wouldn’t say to much about. The next step is to import the photos into software that can render a composite HDR image from multiple shots. Believe it or not, this is the easy part, although the software isn’t free. Photoshop’s latest release, CS5 has an HDR engine. Photomatix Pro is another option, and that’s what I prefer. You can find out all about Photomatix at the website of its parent company, HDRSoft. To blend the images, Photomatix places them on top of one another, aligns them (upon request), and tone maps them. To oversimplify things, tone mapping is like going to each area of the composite picture and saying “which of the original exposures image represents the best exposure of this area of the scene?” The software crunches away, answering that question for all regions of the composite. After its done you have lots of creative freedom to strengthen or weaken the tone mapping effect, as well as many other options. Its actually a lot of fun to play with the controls in these programs to see what kind of looks you can come up with. There are plenty of resources to help you understand how to use Photomatix, and I provide links to a couple at the end of this post. For now, here is my final image of Lincoln Center made from the 8 shots above:

There’s Got to Be an Easier Way

True. An alternative to taking all of the above shots is to take 3 exposure-bracketed shots. Bracketing is a pretty standard function on just about every DSLR and many digital compacts. Using my 5D Mark II, I can take 3 shots in a row that are spaced in exposure according to my settings. For instance, I can set up my camera to take a shot at the right exposure, then take a second shot underexposed by 2 stops, and take a final shot overexposed by 2 stops. The advantage of this approach is that I can take multiple, exposure-bracketed images by pressing and holding the shutter down just once, increasing the chances that I can hold my position for all three shots (and as long as I’m not too far off, Photomatix can usually help me out with fine alignment). So, no tripod necessary (sense a theme?). Here is an example of three bracketed shots that I combined to create an HDR image.

You can see that I spaced my exposures by two stops (the shutter speed is cut by a factor of 4 between shots; each factor of two is one stop). Aside: You can also see that I foolishly shot a landscape shot at f/2.8 on a bright, sunny day. I could have gotten much more depth of field out of this image and would probably go up to f/9 or so if I had another chance. The tradeoff is that the shutter speeds get slower, but not slow enough to introduce any blur from handshake. And here is the final output:

I was relatively happy with the image, but another thing I would change is the people in the image. Off on the right, you can see that there are ghosted images of the people. This is just an artifact of the processing of multiple images with moving subjects. I don’t know enough yet to figure out how to eliminate that using Photomatix, if there is a way at all. I could clone out the ghosts, but what if the moving subjects are key to the scene? That leads me to….

HDR From a Single Exposure

The U.S. Open image above is an example of an HDR image produced from a single exposure. How do I get the other exposures I need? I use Lightroom (or Aperture, or Photoshop, etc). There are several reasons I took this approach. You can probably guess the first reason. No tripod. Second, I want to make sure there is not ghosting. And people move…even while watching tennis. So multiple shots is kind of a non-starter. I also can see that, while the scene does contain a fair bit of dynamic range, its not nearly as much as an extreme example, like the Lincoln Center. This means that, if I shoot RAW, I can probably rely on Lightroom to create version at exposures of +2 and -2 without losing much detail at all. It’s worth restating that shooting in RAW format makes this technique infinitely more effective, because of the additional processing latitude that RAW files offer.

Here is an example of HDR from a single image taken during recent trip to Rocky Mountain National Park:

The only shot I took in this sequence is the middle one. Once I uploaded that image to Lightroom, I used the software to create two additional images: one image overexposed by 1 stop and another image underexposed by 1 stop. Once I loaded the three into Photomatix, I was able to produce an image that was a really close approximation of what my eyes actually saw:

Using the Histogram

When I plan to use a single image to create an HDR composite (because the subject is not perfectly still or I just dont have a tripod), I like to use another trick. I use exposure bracketing, as described above, to help me figure out how many stops apart to space the images and to find the proper exposure. Since my camera can only bracket three images, the goal is to cover the entire dynamic range with those three, even though I’m only going to work with one in post. First, I shoot the three bracketed images using some exposure that feels right. Then I check the histograms for the underexposed and the overexposed images. What am I looking for? I want to get the histograms as close to the sides of the graph without clipping. That means that the histogram gets close to the edge but it doesnt’t run off the right side of the graph in the dimmest exposure…and it doesn’t run off the left side of the graph in the brightest exposure. If both are true, SUCCESS: I have captured the entire dynamic range of the scene in my three exposures. The image below explains what I mean. If there is clipping on one side of the histogram, adjust the exposure to correct it. If both sides are clipped, expand your brackets by 1/3 or 1/2 stop. On the other hand, if you have lots of room on both sides of the graph, contract your bracketed shots by 1/3 or 1/2 stop to spread out the histograms of the bracketed exposures.

With this technique, my camera is basically a diagnostic tool, as I adjust settings until the bracketing setting and exposure are sufficient to cover the dynamic range of the scene. For this patio scene, bracketing at +/- 1 stop was perfect. I can then rely on Lightroom to recreate the dimmer and brighter exposures from the middle one by using the Virtual Copy command and adjusting the exposure slider for each copy. I recommend keeping all three bracketed images. You may need to refer to them to remember how much you bracketed. (Note: these histograms were taken from Lightroom and are of the middle image and my two virtual copies. That’s why the shot settings are all the same.) After I send my middle exposure and my two virtual copies through Photomatix, here is the end result:

This image was part of an architectural shoot, so I removed that stubborn mattress tag while I was at it. This is probably my most frequently used method. I love it especially because I know that alignment will not be an issue at all. Of course, trial and error approaches are not ideal for every scenario. However, for times when you can’t handhold that brightest exposure or want to include people, its works great.

HDR Panoramas

Recently, I have experimented with combining several of these techniques. In July, I created a panoramic scene in Rocky Mountain National Park in HDR. I shot the entire thing hand-held, which wasn’t too terrible, given the fast shutter speeds I used. The image is a merge/stitch of 10 HDR images, shown below.

Each of the 10 HDR image is made up of 5 different exposures. The way I did this was by shooting each of the 10 sections bracketed. For each section, I bracketed at +/- 2 stops. Then, in Lightroom, I created two additional exposures using the Virtual copy command, for a total of 5 exposures spaced 1 stop apart. Here is an example of what I did for one of the 10 sections:

Photomatix was used to generate the 10 HDR sections. I was careful to use the exact same Photomatix settings for each segment to guarantee no problems with exposure, color temp, etc. They were then combined using Photoshop’s merge function. After a few tweaks in Lightroom, here is the final image:

Bringing It All Together

Another one of my favorite images was created by reversing the process I just described. The image is a stitched/merged image with 5 sections. I shot the 5 initial images by panning across the scene in a rainbow shape:

Because I didn’t think I had a great chance at handholding for bracketed shots (and, of course, I had no tripod), I decided to work from single images. Also, the histograms of my bracketed shots told me that this scene covered more dynamic range than my camera’s +/- 2 stops of exposure could capture. So I just tried to use an exposure that would hold as much of the sky detail as I could without leaving the church in complete shadow.

Next, instead of creating HDR images of each segment, I created the stitched image first using Photoshop. Next, 4 Virtual Copies of the stitched image were created:

The distortion was generated by the rainbow motion while panning. I had experimented with this before and love how the buildings seem to stretch skyward from their foundations. It was also right about now that I realized that I wanted to go black and white with this image. I just didn’t like how the sky was going to turn out based on looking at Virtual Copy I. So after Photoamtix completed the HDR image using these five exposures, that’s exactly what I did:

When it comes to capturing images for HDR processing, there are clearly a number of ways to get the job done. I’ve attempted to share a few of the shortcuts that have provided me with good results. These shortcuts are not exactly the same as shooting HDR the long way for many reasons (addendum: See the first comment below for some insight into why; and check out the commenters images…they are some of the best images I’ve ever seen when it comes to using HDR to represent what the human eye actually sees), but they work wonders if you’re not shooting magazine quality images.

Not every scene calls for HDR techniques. In fact, most of what I photograph absolutely doesn’t. The elk image from above is an example of one that may have been fine without it. It does take a bit of practice to be able to determine which scenes deserve the extra effort involved in HDR processing. I’m still working on it. To see the work of a real HDR master, to check out Trey Ratcliff’s blog, Stuck In Customs. It is an entire site dedicated to HDR photography. Trey is both talented and prolific, and his blog offers up many of his secrets to producing some of the best HDR images around. For an absolute treasure chest of Photomatix help, check out the Photomatix Resources Page.