A couple of weeks ago, I took a couple of pictures of the Austin skyline above the trees along the edge of Town Lake. My two best images had different strengths. One had a better contrast of the buildings against the sky, but the details of the tree foliage were lost in darkness. The other had better exposed the shadowy details of the trees but the skyline was not as well defined, being a bit over-exposed.
Today, as I jogged over the MoPac bridge today where I'd taken the photos, I wondered if maybe the best of the dynamic ranges could be combined into a single images with details in both the highlights and the shadows. I'd done some work my freshman year at Duke for a researcher where multiple images were combined to create a 3D image and some of that work, with aligning similar but not identically aligned images, seemed to suggest it was possible.
I found it a fascinating technique. The goal was to make a 3D image of a protein from a series of electron microscope images of the protein that had been processed and attached to a lipid bilayer. The alignment was done by processing the images via some variation of a Fourier transform and the result was a rough grid of points that didn't resemble the original image at all. My job was to use some human pattern-matching skill and intuition to correspond points on the grids of the various images on a computer. Using that data, the computer employed some algorithm to create the 3D image.
While the transform was ultimately and usually performed by computer, it could also be done by shining a red laser through a celluloid negative (film) from the electron microscope and then using some refracting device. The result was the strong points of light that define the same grid that the computer would produce. It could be a bit tricky to use the laser device -- I never quite mastered it that year. A good negative would produce a well-defined, regular grid. Using this technique, the best negatives could be identified. They were then scanned (by me using an amazing high-quality negative scanner) and computer processed.
The good researchers, a husband and wife team, had invited me back to work with them the next year, and were pleased enough with my work to suggest that if I wished to continue from year to year, I could probably join in the research effort in time and get my name on a paper or two. Unfortunately, the Univ. of Florida gave them a good offer and they moved their research down to Florida before I returned for my sophomore year.
Getting back to the topic of combining images to maximize well-exposed detail, imagine my surprise a couple of hours later when I saw this on MacInTouch:
Photomatix Pro 2.0 extends the dynamic range of digital photos or scans by combining images taken with different exposures into a single new image. It offers five different methods for computing the optimal combination of aligned images, 16-bit support, an alignment tool for out-of-register images, and more. The new version adds generation of an HDR image from 24-bit photos, an HDR viewer, and a Tone Mapping tool for scenes with a particularly high dynamic range. Photomatix Pro is $99 for Mac OS X and Windows.
Perfect! Well, almost. I'm not sure that it's worth $99 dollars to me. But I'm glad that the software is readily available.