POTW #4: An experiment in HDRI
Posted by Brian.
|
 |
Finally, a POTW that isn’t a macro of a small animal. This photo shows the UCC church across the street from Wilder. Briefly, this scene had a very wide dynamic range, but the dynamic range that can be captured in a single jpeg is low. To retain detail in highlights and shadows, I took 3 exposures (left, click to view larger), all the same aperture (f/8) but different shutterspeeds (1/100, 1/25, 1/6 sec) and merged them into a single image.
A more detailed explanation:
High Dynamic Range Imaging (HDRI) is a way of increasing the range of luminosities represented in an image. The static contrast ratio of the human eye is around 100:1 (according to wikipedia), which corresponds to around 6 and 1/2 stops. This means that if you look at a scene without moving your eyeballs, you can simultaneously resolve objects whose luminosities differ by a factor of around 100. The dynamic contrast ratio of the eye is a much more impressive 20 stops (or about a 1,000,000:1 ratio of luminosities). That means if you move your eyes around, they can adapt to see a much wider range of luminosities (just not all at the same time).
My camera, however, can only resolve a modest 5 stops in a single scene (stored as an 8-bit per color channel jpeg; this post really deserves its geek tag, doesn’t it?). Most cameras have a similar limitation. Consequently, when you look at a photo taken with basically any camera (digital or film), and displayed on a typical monitor or on photo paper, the luminosity information has been truncated, possibly a lot.
One way to convey more of the luminosity information from the original scene is by combining multiple exposures and compressing the luminosity data to a range that can be displayed in a single scene. In the example above, I’ve taken 3 photos, each separated by 2 stops and combined them to yield a single photo that retains local contrast information in both the highlights and the shadows. It’s important to note that while the local contrast information has been better retained everywhere in the image, the total dynamic range has not been increased and is still limited by the maximum dynamic ranges of the file format and the display device.
August 27th, 2007 at 9:07 am Using
That’s good stuff, guy. And pretty geeky… for the next one you can cover wavelets and discuss jpeg compression algorithms?
It looks as though you can create posts in the future, but once you comment on them, those comments become visible b/c they are made “now” and not in the future. I think you should either (a) be able to change the date of the comment (I’ll try that as soon as I post this) or (b) remove the comment and instead use the “more” html tag?
August 27th, 2007 at 1:25 pm Using
I had several friends and collegues who were all into the wavelets. Let me know if you want the hook-up; I can definitely put you in touch with folks who will have you up to your eyeballs in wavelets…
August 27th, 2007 at 3:22 pm Using
How did you actually make your HDR image? There was an article in Photo Techniques magazine this month or last about how to make them using photoshop.
It’s kind of crazy that it looks like the leaves are starting to change since it’s still in the 90s here. I was going to ask if the blown-out nature of the red leaves was a side effect of the process too, but I just looked at the picture with a different monitor and I think the problem is calibration on my end.
August 27th, 2007 at 4:26 pm Using
I think the red leaves in the tonemapped HDR actually look more like the leaves did on the tree than the red leaves in the individual LDR exposures. I made the HDR image and tonemapped it using Photomatix Pro. I then made slight adjustments to levels and curves in GIMP. Leaves are not an ideal subject for HDR because the multiple exposures are not taken simulaneously. I think that might have something to do with most of the artifacts in the leaves.
I found this tutorial to be useful and informative. Several sources seem to be more satisfied with the results from Photomatix Pro than those obtained using Photoshop.