Spotting a fake photo is a careful exercise in recognizing the details — and even then, some images can fool even the experts. But what if the process of weeding out faked photos started from the source, the camera? That’s the idea behind a new study from the New York University Tandon School of Engineering.
During the study, the researchers considered whether implementing artificial intelligence in the camera could help additional machine learning programs better identify if the image was altered. The researchers found that the method increased a computer’s accuracy at spotting fakes from 45% to 90%.
The technique involves reimagining the way a camera processes information to create a photograph. The camera essentially creates intentional, unique artifacts in the image that software can later use to gauge whether the photograph was manipulated or not. Unlike earlier methods, the software isn’t confused if an image is downsized, the researchers said. When added to the image itself, artificial intelligence programs for spotting manipulated programs were able to boost their accuracy over previous methods.
The tradeoff, however, is that those artifacts, which serve as a sort of digital watermark, degrade image quality. Camera and lens manufacturers continually work to eliminate artifacts, while the researchers intentionally built them back in. The group said that future studies could look at ways to integrate the digital watermark without affecting the quality of the image as heavily.
In order to use the process, camera manufacturers would have to agree to implement the artifacts into their equipment — a task that will be tough if researchers don’t find a way to preserve the image quality. With close competition between different camera models and photographers expecting increasing quality with every new model, intentionally adding artifacts would feel like a step backward for many.
Identifying fakes using software rather than a human eye could help diminish the spread of fake photos, however. Social media networks, for example, could integrate the software to reduce the spread of fake photographs and fake news. However, the technology is considered something that could help experts spot fakes, not something that would help the average person spot a fake photo.
- Image-recognition A.I. has a big weakness. This could be the solution
- Exoskeletons with autopilot: A peek at the near future of wearable robotics
- Summon this amazing Twitter bot to colorize your black-and-white photos
- Facebook’s new image-recognition A.I. is trained on 1 billion Instagram photos
- Evolving, self-replicating robots are here — but don’t worry about an uprising