Page 24 - Detecting deepfakes and generative AI: Report on standards for AI watermarking and multimedia authenticity workshop
P. 24
Detecting deepfakes and generative AI: Report on standards for AI
watermarking and multimedia authenticity workshop
• Robustness with the extraction of watermarking not being affected by moderate cropping,
cropping fill, pixel replacement, image compression, zooming, rotation, noise, blurring,
combination, and other interferences.
• Invisible watermark that currently supports the watermarking of bitmap images, vector
graphics, videos, and audio.
Figure 11: Comparison between SenseTime’s solution and traditional watermarking
methods
Jon Geater, Chief Product and Technology Officer at DataTrails and Co-Chair of the IETF
SCITT Working Group, mentioned in his presentation that all work on relevant regulations and
standards is pointing towards durable, high-integrity accountability, transparency, and data
provenance for AI-generated content.
Figure 12: Trust elements
16
