Deepfake Geography: How AI Fakes Satellite Images

With deepfake technology, it’s possible to fake not only video images of faces to bypass biometric authentication systems, but also satellite photos of landscapes. This has scientists concerned that so-called “deepfake geography” could become a major problem in the near future.

In a study called “location spoofing,” researchers at the University of Washington used photos of three cities and various image and video manipulation techniques to develop new ways to detect fake satellite images, warn about the dangers of falsified geospatial data, and call for the creation of a system to verify geographic facts.

“This isn’t just Photoshop. It’s a way to make data incredibly realistic. These techniques already exist. We’re just trying to identify the potential uses of these techniques and develop a strategy for what to do about them,” explained Bo Zhao, associate professor of geography at the University of Washington and lead author of the study.

Zhao and his co-authors note that fake locations and other inaccuracies have been used in mapmaking since ancient times. Part of this is due to the very nature of transforming real locations into map form, since no map can represent a place exactly as it is. But some inaccuracies in maps are deliberate fakes, intentionally created by cartographers. There’s even a term, “paper towns,” which refers to fake cities, mountains, rivers, or other features quietly placed on a map to prevent copyright infringement.

With the rise of geographic information systems using satellite imagery, like Google Earth, location spoofing has become much more sophisticated and risky.

To study methods for creating fake satellite images, Zhao and his team turned to an artificial intelligence framework previously used to manipulate other types of digital files. In the context of mapping, the algorithm closely studied the features of satellite images of urban areas, then generated deepfake images by transferring the characteristics of one satellite image onto a different base map—similar to how popular filters can put human facial features onto a cat’s image.

The researchers then combined maps and satellite images of three cities—Tacoma, Seattle, and Beijing—to compare their features and create new images of one city based on the characteristics of the other two. They chose Tacoma as the base map. By taking the geographic features and urban structures of Seattle (which is similar in topography and land use) and Beijing (which is different in topography and land use), the researchers merged them to create deepfake images of Tacoma.

Source

• Our other channels
• Our friends and partners