Science

Princeton Computer Scientists Want to Give You Spellcheck for Your Photos

Adios, trash, and photobombers. 

by Ben Guarino
Flickr.com/Buie

If you care about such things as aesthetics, perhaps you are of the mind we are drowning in bad photos. On Facebook alone, people upload 350 million images a day. Princeton University computer scientists and Adobe researchers are fed up with your casual, carefree approach to documenting life events. And rather than teaching you how to compose a photo based on the rule of thirds, because art is dead, or any sort of post-image editing software, they want to simply create an algorithm that makes the images better.

This isn’t an Instagram filter that paints over the entire photo in a glossy coat, you hack — the scientists want to target and automatically remove what they’re calling “distractors.” Think of it as a sort of spellcheck for your snapshots, when you or the elements fuck up the grammar of composition.

Ohad Fried and Adam Finkelstein, Department of Computer Science, Princeton University; and Eli Shechtman and Dan Goldman, Adobe Research

The researchers outline their methods in a new paper, “Finding Distractors In Images,” presented at this year’s Computer Vision and Pattern Recognition Conference. Using Amazon’s Mechanical Turk and an Adobe app called Fixel, the researchers recruited a set of more than 6,000 images that humans pointed out had strange or unwanted elements. Based on these data, the scientists developed an automatic detector for things like cars or faces at the edge of images — essentially photobombs, inadvertent or otherwise — and it smooths over the flaws, akin to automatic portrait retouchers.

The system, like every photo-recognition or detection software to precede it, has problems (sometimes the detector got rid of parts of the photo that you want to keep). Ohad Fried, a Princeton graduate student and the paper’s lead author, is optimistic that it will become accurate enough to go to market.

If the service ends up as simple as adding an Instagram filter, you can expect our fictionalized Facebook documentaries to get even rosier. Should we care? The $4.3 million digital bleakness of Rhine II says it’s fine. But if you believe that photos should be as honest as quotes, is it OK to touch up less-than-elegant grammar? Blemishes are a fact of life; to CGI them out of existence is a fiction. If you’re a stickler for the truth, just reframe your shot.