To ensure we never miss a moment, it's entirely plausible that one day cameras will just continuously record video, instead of snapping individual shots. But when all you want is a single photo, pouring through hundreds of frames in a clip for that perfect shot is overly tedious. So researchers have come up with a better approach to automatically analyzing videos and selecting only the best candid portraits.

Since there's no scientific formula for determining what constitutes an ideal portrait, researchers at the University of Washington, working with engineers from Adobe, first started with human test subjects rating frames from multiple videos based on the aesthetics of the image. Their results were then used to create a computer model that weighed several different features to determine if the subject's face, in a given frame, represented an ideal portrait of that moment. Previous techniques took a simpler approach by omitting frames in a video where the subject blinked, was in motion, or even just smiling, which still resulted in selections that looked like the awkward in-between faces seen in a paused video.

The research actually seems like a solution to a problem that doesn't quite exist yet, but immediate applications include improving the thumbnail selection process for video sites like YouTube, or automatically grabbing better shots of a suspect's face in security camera footage. [University of Washington via John Nack]