Adobe's new AI can identify altered images
Recent announcements from Adobe have suggested that future updates to digital tools such as Photoshop will give users an unprecedented ability to manipulate images. In a rare case of corporate awareness of social responsibility, the company has just revealed it is working on an artificial intelligence system that can quickly detect whether an image has been artificially manipulated.
The old adage of "a picture is worth a thousand words" arose in the early 20th century, often in the context of a newspaper article. The idea was that an image can represent the truth of a situation better than any description. Of course, over the last decade that old adage has come under attack from modern technology and could now be almost rephrased as "a picture is worth a thousand lies."
Adobe has been at the forefront of democratizing image manipulation technology. As it has constantly updated its Photoshop software, it has become easier and easier to create photorealistic images that seem entirely authentic. Last year, the company revealed even more upcoming tools that are brilliantly useful, but also further erode any trust one can have over the authenticity of an image.
While there are currently a variety of forensic techniques that allow experts to identify whether an image has been artificially manipulated, they are often time-consuming and limited in their scope. Vlad Morariu, a senior research scientist at Adobe, set out to harness artificial intelligence and machine learning to help create a tool that can quickly and accurately detect whether an image has been digitally modified.
"We focused on three common tampering techniques – splicing, where parts of two different images are combined; copy-move, where objects in a photograph are moved or cloned from one place to another; and removal, where an object is removed from a photograph, and filled-in," explains Morariu.
All of these manipulation techniques leave specific artifacts on an image but these distortions are not easily visible to the human eye. A digital forensic specialist may take hours to examine an image to determine its authenticity, looking closely on a pixel level for signs of tampering. Morariu developed a neural network that can deploy two deep photographic analysis techniques and quickly determine if the image is genuine or not.
"Using tens of thousands of examples of known, manipulated images, we successfully trained a deep learning neural network to recognize image manipulation, fusing two distinct techniques together in one network to benefit from their complementary detection capabilities," says Morariu.
At this stage, the algorithm is relatively limited to analyzing an image using just two forensic techniques, but there is no doubt that this could be expanded to more techniques in the future. Whether this AI can be easily outsmarted by a devious image retoucher is yet to be seen, but the battle for truth and authenticity is certainly kicking up into new terrain.
The line between real and fake is rapidly being harder to distinguish, but it is vitally important that some kind of ability to track the veracity of an image is available to all. There is no word from Adobe on whether a photo analysis algorithm utilizing this technology will be incorporated into the company's products in the future, but at the very least it is helpful to know the company is aware of the dangerous world its products are creating.
"It's important to develop technology responsibly, but ultimately these technologies are created in service to society," says Jon Brandt, director for Adobe Research. "Consequently, we all share the responsibility to address potential negative impacts of new technologies through changes to our social institutions and conventions."
Take a closer look at the new AI tool in the video below.
Source: Adobe Blog