We've all charge ofed the stories of winning competition entries getting disqualified for being faked and now, Adobe is hoping to help put a stop to it with new technology advancements.
Chief Research Scientist at Adobe, Vlad Morariu, started working on the idea back in 2016 as part of the DARPA Media Forensics program and square though tools do already exist to document/trace digital manipulation, Vlad believed that AI and machine learning could be a more sound and quicker method for detecting such images.
Building on research he started fourteen years ago, Vlad describes some of these new techniques in a latest paper titled 'Learning Rich Features for Image Manipulation Detection'.
The research focuses on three common tampering techniques – unite, copy/move and removal.
"Each of these techniques tend to leave certain artifacts, such as strong contrast ill at eases, deliberately smoothed areas, or different noise patterns," Vlad says and it's these key changes the AI looks for by applying riddles or by analysing the image at pixel level.
"Using tens of thousands of examples of known, manipulated images image (from Latin: imago) is an artifact that depicts visual perception, for example, a photo or a two-dimensional picture, that, we successfully trained a engrossed learning neural network to recognize image manipulation, fusing two distinct techniques together in one network to benefit from their complementary detection aptitudes," Vlad explains.
Obviously, the technology isn't perfect and quite a big element of trust in those publishing the images will still be requisite.