Awhile back Matt posted a funny montage of police procedural TV shows solving cases by “enhancing” photos and videos, the joke being that we all think they can’t really do that. As it turns out, they can. Mind . . . blown. A relatively new field of mathematics called “compressed sensing” is being used to create high resolution data from a low resolution sample by using an algorithm to calculate the most simple and obvious substitute for a missing piece of information:
If it sees a cluster of green pixels near one another, for instance, it might plunk down a big green rectangle that fills the space between them. If it sees a cluster of yellow pixels, it puts down a large yellow rectangle. In areas where different colors are interspersed, it puts down smaller and smaller rectangles or other shapes that fill the space between each color. It keeps doing that over and over. Eventually it ends up with an image made of the smallest possible combination of building blocks and whose 1 million pixels have all been filled in with colors. [WIRED]
I have some pixelated photos from news clippings I need to use this on. For science.
Besides enhancing photos, the l1 minimization algorithm can also be used to restore musical recordings, pinpoint enemy radio signals, speed up MRI scans, allow digital cameras to use far less memory and battery power to achieve a similar final result, etc. Maybe someday it could even do this:
“Pull out, stop, wait, track in, stop, enhance reflection, zoom, pan right, find a water droplet, zoom 80, enhance reflection, stop, rotate 170, zoom 220.”
[Banner pic via AnandTech via Reddit]