Megapixels, Shmegapixels: How to Make Great GIGAPIXEL Images With Your Humble Digital Camera
Size matters, at least when it comes to the resolution of digital photos. As much as I love my iPhone, its built-in 1200×1600-pixel camera just doesn’t work for serious photography. Problem is, it’s been my only camera for almost a year, ever since my previous digicam croaked during a cross-country road trip. But a couple of weeks ago I found a very reasonable price ($314) on a Canon Powershot S5, which has maximum resolution of 3264×2448 pixels, or 8 megapixels. That’s enough to make nice 16-by-20-inch prints, and should be plenty, practically speaking, for anyone but a professional photographer.
Or so I thought. But I’ve recently become intrigued with a form of digital photography that takes image size to a new extreme: super-high-resolution or “gigapixel” imaging. Gigapixel photography isn’t about making bigger prints—it’s about gawking at the images themselves on the screen, sort of the way you’d watch hummingbirds fly in super-slow-motion in a Discovery Channel show on your HDTV. Such images contain such an overwhelming amount of detail that you’re not really supposed to download or view them all at once—rather, you use a combination of Web streaming technology and scrolling and zooming tools to dive deeper and deeper, in much the same way that software like Google Earth allows you to zoom from a view of the entire planet to a view of your backyard.
Of course, there’s no commercially available camera that can actually take pictures with billions of pixels. Hasselblad’s H2D-39, a $31,000 device based on Kodak’s 39-megapixel CCD, represents the rough upper limit today. To make images bigger than that, you have to stitch multiple photos together using graphics software. And lately, two things have happened to make gigapixel photography a practical pastime for amateur photographers: professional-grade stitching software has become more affordable, and there’s at least one online community, GigaPan.org, where you can upload and share your gigapixel images.
I mentioned the GigaPan site briefly in my first World Wide Wade column back in April. It’s not a Flickr-style commercial photo-sharing site, but rather an outgrowth of a collaborative research project between Google, Carnegie Mellon University, the Intelligent Robotics Group at NASA Ames Research Center, and an Austin, TX, hardware maker called Charmed Labs. This fall Charmed is bringing out the Gigapan Robotic Imager, a motorized platform for consumer digital cameras. The device precisely controls a camera’s pan and tilt, guiding it through dozens or hundreds of overlapping snapshots that can be stitched together later to create huge panoramas or mosaics. (NASA probes such as the Mars Phoenix lander generate panoramas of the Martian landscape in much the same way.)
But you don’t need a fancy device like the Gigapan robot to make great panoramic photos. And while GigaPan.org was originally developed as a showcase for images created using the Charmed Labs imager, anyone is free to upload their panoramas to the site. There’s just one catch: unlike other photo-sharing sites such as Flickr, which has an upper limit of 20 megabytes on the size of the photos you can upload, GigaPan has a lower limit: images must be at least 50 megabytes in size!
As a weekend project, I decided to see what kinds of panoramas and mosaics I could make for the GigaPan site using my new Canon. I grabbed my tripod and trooped over to Boston’s Copley Square, home to architectural gems such as Trinity Church, the John Hancock Tower, the Boston Public Library, and Old South Church.
Like most Powershot models, the S5 has a built-in “stitch assist” mode that helps you take a series of images from left to right. In other words, it shows the rightmost edge of your previous photograph on the screen to help you line up the next shot, theoretically leaving enough overlap so that photo-stitching software can later create a seamless panorama. Notice I said “theoretically.” I used stitch assist mode to take 13 successive images of Copley Square, but when I got home and used the Canon PhotoStitch software that came with my camera to assemble the images, the resulting image (click on the link or on the thumbnail above to see it) was less than ideal. Not only did the software have a tough time lining up the images correctly, but the final panorama suffered from optical distortions that made most of the buildings in the photo lean precariously, as if gravity had gone askew.
It wasn’t really the software’s fault—it just wasn’t made to deal with a particular challenge in photography called the “keystone effect.” If you’re standing on a sidewalk and looking up at a building, perspective will naturally cause the higher floors of the building to look narrower. Our brains seem to correct for this effect most of the time, so we don’t really notice it. But photographs have a way of calling attention to it, and one result is that a tall building (or any vertical edge) that isn’t near the center of an image will appear to lean perilously toward the center line. Since a panorama consists of a series of individual photographs—each one keystoned around its own center line—the collection as a whole can be vertigo-inducing, with buildings tilting every which way. To fix the problem, you need better software.
So I downloaded the trial version of a program called PTGui (for Panorama Tools Graphical User Interface), one of the third-party products recommended by GigaPan.org. (The Gigapan device will come with its own stitcher software.) Made by a small software company in the Netherlands called New House Internet Services, PTGui includes a mess of clever features that correct problems like keystoning. I don’t really know how it works, but I’m guessing that the software has edge-detection algorithms that help it select the portions of each image that are least affected by keystoning; it may also rotate individual photos as necessary to make vertical lines look vertical. To see how well it works, just take a gander at the alternate panorama I made with PTGui, using the same source images as before. In this image, the buildings are as straight as pine trees.
I also wanted to try making a large mosaic. So I positioned myself on a Copley Square sidewalk kitty-corner from Old South Church and used a telephoto setting to take 42 extreme close-ups of the church, a lovely nineteenth-century Northern Italian Gothic structure built from local Roxbury puddingstone. To put PTGui’s image-recognition capabilities to the test, I dumped all 42 pictures into the program at once and let the software sort out how they fit together. It did so without difficulty, with the exception of a couple of tiles that showed nothing but sky and clouds, which I had to reposition manually.
The resulting mosaic, of course, suffered from the same keystoning as any single image would have—but one of PTGui’s marvelous features is that you can drag the entire mosaic around vertical or horizontal control lines, which has the effect of warping the appropriate areas of the mosaic until all of the vertical or horizontal edges are parallel. The result is a mathematical fiction, but somehow looks much more pleasing to the eye. To see what I mean, look at these views of my mosaic before and after I applied the vertical-line correction.
I haven’t yet put my full Copley Square panoramas and Old South Church mosaic up on the Web, because the trial version of PTGui doesn’t allow you to save images. (I captured the images you see here directly from my computer screen using the Mac’s Grab utility.) I had almost made up my mind to plunk down the 79 euros ($126) that New House Internet Services charges for the full version of the program when the PTGui website went offline. (I wonder if they’re hosted by The Planet?)
I’m sure I’ll buy the software, or some similar program, pretty soon, as I can feel an itch developing to be part of the community of hundreds of gigapixel photographers who are sharing, exploring, rating, and annotating one another’s images at GigaPan. Travel- and nature-related photos seem to be the most popular: indeed, this 240-megapixel image of the Grand Canyon demonstrates just how much detail can be packed into a mosaic (you can see ripples on the Colorado River from miles away). One of the coolest features of the GigaPan site is the ability to isolate “snapshots” in other people’s mosaics and post them alongside the main image, almost as if they were comments on a blog post; the feature frequently leads to a sort of game in which site users race each other to point out funny, obscure, or surprising details that aren’t visible until you zoom way in, the same way Google Earth users collect instances of nude sunbathers on rooftops.
Given Moore’s Law, it wouldn’t be a shock to see gigapixel image sensors turning up eventually in consumer-level cameras. But until that day, super-high-resolution photographers will have to stick to the software equivalent of the quilting bee, stitching individual panels together into massive panoramas or mosaics that only appear as if they were unified images. The reward can be a picture so finely featured that you feel like you can keep zooming into it forever, as if you had Superman’s telescopic vision. Now that’s high definition.
Update 6/6/08: Related to gigapixel imaging is the technology of Deep Zoom being developed by Microsoft Live Labs, using its Silverlight and Seadragon technologies. See this nice PC Pro article and the Hard Rock Cafe demo it references (you’ll need to install Silverlight on your computer to make it work).