KAIST develops deepfake detection tool

Korea Advanced Institute of Science and Technology has developed an artificial intelligence-powered tool designed to detect computer-manipulated images, the institution said Tuesday.

The new Kaicatch software, developed by the KAIST research team led by professor Lee Heung-kyu, is the nation’s first deepfake detection tool, the institution said. The software is capable of analyzing both images and videos to find artificial changes that might be invisible to the human eye.

The team developed the software to counter the growing number of synthetic photos and videos that could pose potential problems. The institution expressed hopes that the new software helps vouch for an image’s validity.

To build the new detection tool, the research team combined various techniques, including those used for video forensic and steganography analysis. It has used a public data set of around 300,000 images and a large amount of video data to develop the tool, the team said.

The tool utilizes two different machine learning engines that are designed to look for areas that have been altered to figure out the types of techniques used to manipulate images.

Professor Lee said his research team will improve the software’s capabilities to detect sophisticated changes made to photos and videos going forward. 

By Shim Woo-hyun (ws@heraldcorp.com)