Obstruction-Free Photo Algorithm

An Algorithm Could Help Remove Some Objects From Photos

August 6, 2015 - Written By Diego Macias

Cameras have become integral components for smartphones, not only do they offer better quality than some point-and-shoot cameras, but they can be used to scan documents with additional software or recognize QR codes. Some advancements in this department include the ability to capture larger pictures so less detail is lost while zooming in and OIS along faster autofocus can help to capture sharper images even with moving objects. Also, lenses are now wider, so more elements can fit into the picture and low-light photography has been improved with the aperture of the new lenses. Some of these technologies have even made their way into the front-facing cameras.

Software on the camera apps has also been vastly improved and now some tricks like changing the focus after the photo was taken, adding AR objects or some geometric effects are now possible. Even removing some objects from the picture is possible, as the camera captures a series of pictures and uses some parts of them in order achieve the effect. Now, researchers at Google and MIT have created an algorithm to remove some obstructions caused by reflections, raindrops or other thin objects like a fence. It works in a similar way, as the camera captures a series of pictures, but this time they’re captured from different angles and thus, being able to remove the undesired objects.

The video provides various samples and conditions on how this technology would work and it seems successful in most cases. Of course, there are some areas that need improvement, but it is clear where this technology is going and it looks promising. Even more impressive is the way the algorithm manages to recreate the objects from their reflections. This works for various textures and under several light conditions. Some limitations of this technology is that it won’t work with moving objects or if it has to deal with two layers of objects, for example trying to take a picture from inside a car to an object behind of a fence. It’s not clear when the algorithm will be released to be used in commercial devices, but since they use a smartphone to demonstrate its functionality, it becomes clear that current technology could easily integrate it.