It is being trained with manipulated images so it can learn how to spot photos that have been edited
In today’s digital world, it has become quite easy to transform a completely dull picture into a piece of art, which in a way, is a devil in disguise. With social media and instant photo sharing platforms going all out, it has become way more vital to debunk ‘doctored’ photos. Not only do they encourage wrong perception among the audience but also lead to fake news, hate speech, religious quarrels and so on.
To counter the issue of fake photo manipulation, Adobe is said to be workingon an advanced artificial intelligence system. The system has a neural network that is being trained with manipulated images so it can learn how to spot photos that have been edited.
According to Adobe researcher Vlad Morariu the project is a part of the government-sponsored DARPA Media Forensics program. There are already certain tools that can help in detecting photo manipulation, for instance one can look at the metadata and certain programs can examine features of photos like noise, edges, lighting, pixels to figure out if the photo has been edited.
Artificial intelligence is taking things to the next level. During the research, the team focused on three common image manipulation techniques, ‘splicing’ where parts of two photos are combined, ‘copy-move’ where objects in photo are moved/cloned from one place to another, and ‘removal’ where objects removed from photos and filled in.
“Each of these techniques tend to leave certain artifacts, such as strong contrast edges, deliberately smoothed areas, or different noise patterns. Using tens of thousands of examples of known, manipulated images, we successfully trained a deep learning neural network to recognize image manipulation,” says ,” Morariu.
Basically it analyses the noise profile of an image, and determines whether any given part of the photo has a different pattern. If there is a break in pattern in any section of an image it can be deduced as spliced. It also checks for tampering by checking the artifacts which can include imperfect edges, contrast levels and so on. While there is no plan as of yet to make this software open for public, it could end up as a forensic tool.