Sign In

Communications of the ACM

ACM TechNews

AI Removes Unwanted Objects From Photos


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
A photo of a tiger from which the algorithm has partially deleted the fence.

Virginia Polytechnic Institute and State University researchers developed an algorithm that uses artificial intelligence to digitally remove unwanted objects from photos.

Credit: Xue, Tianfan, Michael Rubinstein, Ce Liu, and William T. Freeman

Researchers at Virginia Polytechnic Institute and State University developed an algorithm that uses artificial intelligence to digitally remove unwanted objects in photos.

The neural network analyzes several frames in a motion photo taken by a smartphone, identifies various objects in the image, and calculates the distance to each object.

It then separates the objects into different layers, removing the foreground layer to provide an unobstructed view of the objects behind.

The neural network automatically identifies distracting objects, including reflections, in the foreground of the image during the process of learning.

The algorithm works faster than prior methods, taking just three minutes to decipher a 1,296-by-864-pixel image.

From New Scientist
View Full Article - May Require Paid Subscription

 

Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA


 

No entries found