advertisement
Fake news is an ongoing phenomenon that is one of the most daunting, complicated issues society is facing these days, and it has even proved fatal. Morphed images are a big part of that.
Kaluram Bachanram, a resident of Bangalore was lynched by a mob after WhastApp rumours convinced people that he was a child kidnapper. Kalu became a victim of the fake news firestorm in India. Across India, eight lives have been lost in this latest wave of spreading fake facts.
In the times of Internet, new technologies have opened up avenues for fake photos and videos. However, people in the tech and media circles have already started coming up with ways to fight fake news.
In May, Facebook announced it will take drastic measures to curb fake news. It said that it was making investments to stop the spread of false news and to promote high-quality journalism and news literacy.
Facebook product manager said the new approach will help people to "stay informed and diametrically reduce the reach of false stories”.
Google will also launch a new training programme for journalists to protect them from falling prey to fake news stories. it will train 8,000 journalists in English and six other Indian languages over the next one year, reports IANS.
American multinational computer software company Adobe, known for its multimedia software (especially Photoshop and Premier Pro) has also geared up to take on fake news. Adobe, however, will take the help of Artificial Intelligence to detect fake images.
Adobe's vision is that digital forensics can be used to detect tampered images. According to Adobe, distinguishing authentic images from tampered images has become increasingly challenging.
Adobe will use image manipulation detection by exploring both RGB image content and image noise features, in order to identify tampered images. What this means is that Adobe has proposed a two-stream manipulation detection framework, which not only models visual tampering artifacts (eg, tampered artifacts near manipulated edges), but also captures inconsistencies in image noise.
Adobe has published a research paper showing how machine learning can be used to identify the three most common types of image manipulation - image splicing copies regions from an authentic image and pastes them to other images, copy-move copies and pastes regions within the same image, and removal eliminates regions from an authentic image followed by an edit.
Adobe's machine learning was a result of feeding a large dataset of edited images into the software.
Although Adobe's work towards identifying doctored pictures is not the first attempt at identifying doctored images. Recent work on image forensics uses clues such as local noise features and camera filter array, but those methods focus only on a single tampering technique. Here, Adobe proposes a novel two-stream manipulation detection framework, which not only models visual tampering artifacts (e.g., tampered artifacts near manipulated edges), but also captures inconsistencies in local noise features.
Although this is still a long shot, this research work by Adobe is still a vision. It is good to see work being done to help battle fake news and digital fakes.
(At The Quint, we question everything. Play an active role in shaping our journalism by becoming a member today.)