Less is More With Compressed Scanning

For those who continue to believe that bigger is better—that you’re better off, for example, the more megapixels your digital camera delivers—a recent article by Jordan Ellenberg in WIRED magazine suggests the opposite may be true. It turns out that compressed sensing, a technique discovered in 2004 and that uses algorithms to predict the likely detailed information to be found in an image, is revolutionizing medical diagnostic imaging, aerospace, military software, surveillance, and will ultimately make all sorts of the photographs that are made easier to gather, interpret, and store.

The WIRED article begins by telling a touching story—how minimal data from a squirmy 2-year-old transplant patient’s sketchy MRI was boosted through compressed sensing to extract the vital and detailed visual information that was necessary to save his life. What compressed sensing does is analyze sparse or incomplete existing data to predict what a sharper and more detailed version of that data would look like. The key to the process is what’s called “sparsity” a mathematical algorithm that takes the sparse information in an image and, bit by bit, fills in the blank spaces with what computer programs suggest is the most likely missing information. Far more complex than something like the sharpening programs in Photoshop, compressed scanning can take hours upon hours of computing time, but the results have proved to be consistently and startlingly accurate.

The implications are enormous. Emmanuel Candès, who discovered the phenomenon of CS and now works at Stanford University, believes that that collecting, compressing, and storing vast amounts of visual information may be a gigantic waste of time and resources. Whether you’re taking snapshots of your kids, or are responsible for downloading detailed images from a battery-guzzling camera on a satellite orbiting Jupiter, it will become increasingly practical to gather meaningful data and construct complex imagery from the smallest samplings of information.

Just as importantly, compressed sensing may play a central role in helping us cope with handling and accessing the massive number of visual images we save and, from time to time, need to retrieve. At present, the images and visual information we archive not only have to be saved, but converted whenever the storage formats they’ve been captured with become obsolete. With CS imaging, Candès research suggests, it will only be necessary to record and maintain about 20 percent of the pixels in certain images to be able to accurately reconstitute them later. As collecting institutions, like the Smithsonian, grapple with the costs and responsibility of preserving massive numbers of images now and in the future, compressed scanning suggests a solution for making more from less.