@ericyocam6Professor, California Polytechnic State University
Shipped: Modified Media Detection (DeepFake)
A patented system for detecting manipulated and DeepFake media in a wireless carrier network. U.S. Patents 11,514,342 and 12,086,727.
deepfakemedia-forensicsai-securitypatentcomputer-vision
Comments2
Detecting manipulated media inside the carrier network is a different vantage point from on-device checks, since you see the transport layer and re-encoding artifacts. How do you keep detection reliable after the platform transcodes the video, and does the network position help or hurt that?
Great question. Transcoding and compression can remove or distort artifacts used in deepfake detection, so relying solely on visual artifacts is problematic. Our patented approach uses machine learning to analyze media content alongside additional contextual information, such as device sensor data and location, when available. The network position provides an advantage by allowing detection during transmission, rather than relying exclusively on the receiving device. However, transcoding remains a challenge, and maintaining detection accuracy across different compression levels and codecs is an important area for further evaluation.
2 comments
Detecting manipulated media inside the carrier network is a different vantage point from on-device checks, since you see the transport layer and re-encoding artifacts. How do you keep detection reliable after the platform transcodes the video, and does the network position help or hurt that?
Great question. Transcoding and compression can remove or distort artifacts used in deepfake detection, so relying solely on visual artifacts is problematic. Our patented approach uses machine learning to analyze media content alongside additional contextual information, such as device sensor data and location, when available. The network position provides an advantage by allowing detection during transmission, rather than relying exclusively on the receiving device. However, transcoding remains a challenge, and maintaining detection accuracy across different compression levels and codecs is an important area for further evaluation.