Technology can and does make the world a better place, but some of the tools we use to make positive change can also be turned against us to promote falsehoods. Adobe hopes to help slow the spread of fake news with tools to detect manipulated images. The company worked with UC Berkeley scientists to develop a neural network that can spot fake photos and work backward to restore their original appearance.
Adobe already has some features in Photoshop to prevent people from using them in shady ways. For example, image recognition tools block the manipulation of bank notes. These files won’t even open. Naturally, you can’t prevent people from opening files that include faces, but the new proof-of-concept tool from Adobe and UC Berkeley might be able to tell you if those faces have been altered.
The tool detects the use of Adobe’s Face Aware Liquify tool, which makes it easy to make subtle tweaks to someone’s face in a picture. The detection relies on a Convolutional Neural Network that has been trained with examples of headshots edited with Photoshop as well as the same photos before alteration.
Before testing the accuracy of the neural network, Adobe wanted to know how good (or terrible) humans were at spotting manipulated photos. It showed the altered and unaltered headshots to some human volunteers, asking them to identify which photo was fake. Unfortunately for humanity, they were only able to pick the phony photo 53 percent of the time, and that’s with both images in front of them. The AI trained on manipulated photos was able to spot the fake with 99 percent accuracy.
The detection tool highlights areas of the image that appear to be modified, and it can go a step further. Since the training is based on an established Adobe tool, the AI can reverse the edits to produce a very good approximation of the original photo.
This is an impressive first step in battling misinformation around the internet, but there are uncountable ways to create fake photos. This tool won’t be able to detect anything other than the Photoshop face liquify filter. It’s going to take much more work to create an AI that can account for all or even most image manipulation techniques. That’s probably a good avenue of research, though, seeing as humans are so bad at spotting fakes.