MIT researchers develop algorithm to remove reflections from photos shot through glass

by Gannon Burgett

posted Wednesday, May 13, 2015 at 1:04 PM EST


When taking photographs through glass, there’s almost always the problem of reflections. Of course, you could avoid shooting through glass in the first place, but that’s not always possible.

MIT researchers are looking to rid the world of unwanted reflections though, with a new algorithm that’s capable of removing reflections from an image captured through glass.

The algorithm is based on the premise that almost all photographs shot through double-paned glass have two separate images: one of the subject matter actually being photographed and one of the unwanted reflection.

It breaks down these two images and attempts to remove unwanted artifacts from each, leaving you with two separate photographs. The results from the algorithm aren’t perfect, but they’re a solid improvement over the original input.

The report notes that as more photographs are put through the algorithm, it can better learn to separate the two images, leading to better results down the road. Yoav Schechner, a professor of electrical engineering at Technion, says while this could be used in conventional photography, it also presents an opportunity for image-based recognition of objects by computers that would otherwise struggle to identify the subject matter.

Image credits: Illustration courtesy of MIT