New Technique by NFI Offers Insight into Deepfake Detection
Amsterdam, vrijdag, 23 mei 2025.
The Netherlands Forensic Institute (NFI) has developed an innovative technique to detect deepfakes using blood flow detection. This approach examines subtle colour changes on the face, caused by the heartbeat, which are not present in AI-generated videos. This breakthrough contributes to the fight against disinformation and strengthens trust in visual media, especially in an era where deepfake videos are becoming increasingly indistinguishable from reality. Forensic researcher Zeno Geradts will present these findings next week at a conference in Dublin, marking an important step in the ongoing struggle against media manipulation.
The Rise of Deepfake Detection Technologies
The technologies used to detect deepfakes are rapidly evolving. In addition to the blood flow detection technique from the NFI, there are other methods such as Photo Response Non Uniformity and Electric Network Frequency. Photo Response Non Uniformity analyses the unique ‘fingerprint’ that a camera leaves on each photo or video. Conversely, Electric Network Frequency analyses fluctuations in light frequency to determine where a recording was made[1][2]. These technological advances are necessary because deepfakes are becoming increasingly realistic and harder to detect[2].
Functionality and Effectiveness of Blood Flow Detection
Blood flow detection works by measuring subtle colour changes that occur due to vasodilation with each heartbeat. These variations, present in real footage, are absent in deepfake images. In particular, small veins around the eyes, forehead, and jaw are extremely suitable for this analysis as the blood vessels are close to the skin there[1][3]. While the technique is promising, it is still in the validation phase and is not currently used as standard in legal cases, although it can be used as supporting evidence[3].
The Ongoing Arms Race
The advanced nature of AI technologies results in a continuous ‘cat-and-mouse game’ between creating and detecting deepfakes. AI tools are increasingly adept at simulating characteristics that suggest human authenticity, requiring detection methods to constantly evolve[1]. Despite progress, forensic researcher Zeno Geradts warns that the rapid evolution of technologies requires continuous adaptation to stay ahead of criminals[2]. He states that there is an urgent need for ongoing innovation and research to effectively respond to rapid developments in AI and deepfake creation[3].
Societal Impact and Future Challenges
Deepfakes have become a powerful tool for spreading disinformation and committing identity fraud. The growing ability to create convincing fake videos can significantly undermine public trust in visual media[4]. Geradts expressed concern about the future, wherein no one may believe in the authenticity of images anymore. Combining different detection techniques is deemed crucial to counter these threats to credibility[5].