There is an urgent need for the advancement of IT technology to prevent threats to daily life caused by deepfake videos. Real-time detection of deepfake videos is difficult to implement, so until now, users have had to go through the process of uploading videos to detection applications for analysis and waiting for results for several hours, but this inconvenience is expected to be resolved.

On the 16th, Intel announced that it has developed FakeCatcher technology capable of detecting fake videos with 96% accuracy. This technology was developed as part of Intel's 'Responsible AI' efforts.

Intel's deepfake detection platform is the world's first real-time deepfake detector to provide analysis results within milliseconds. The real-time deepfake detection technology unveiled by Intel runs on a server and interface via a web-based platform using Intel hardware and software.

While most deep learning-based detectors verify the original data to find signs of non-authenticity and identify problems in the video, FakeCatcher obtains clues from the actual video by gauging real human elements, namely the subtle 'blood flow' of humans appearing in the video pixels.

This utilizes the fact that the color of veins changes when the heart pumps blood to collect blood flow signals from human faces appearing in the original video and convert them into spatiotemporal maps using an algorithm. Subsequently, the authenticity of the video can be immediately determined using deep learning.

In terms of software, an optimized fake catcher architecture is configured using various specialized software tools. The development team used OpenVINO™ to run the AI model for face and terrain detection algorithms.

The computer vision block was optimized based on Intel® Integrated Performance Primitives, a multi-threaded software library, and OpenCV, a real-time image and video processing tool; the inference block was optimized with Intel Deep Learning Boost and Intel Advanced Vector Extension 512 (AVX0512); and the media block was optimized with Intel AVX2.

In addition, the development team provided an integrated software stack for the Intel Xeon Scalable processor family using the Open Visual Cloud project.

In terms of hardware, the new deepfake detection platform can run up to 72 different detection streams on 3rd Gen Intel Xeon Scalable processors.

Deceptive acts caused by deepfakes can lead to negative consequences such as a decline in media credibility, and it is expected that FakeCatcher will be utilized in the future for filtering systems when uploading videos and for filtering manipulated video reports by media outlets.

Social media platforms can utilize this technology to prevent users from uploading harmful deepfake videos. Global news media can apply this technology to prevent accidentally reporting manipulated videos. In addition, non-profit organizations can use this platform to make deepfake detection capabilities available for everyone to use.

“You have likely seen videos of famous celebrities doing or saying things they didn’t actually do,” said Ilke Demir, a senior researcher at Intel Labs. “Deepfake videos are now available everywhere.”