AI Content Marking in Video Bitstreams via SEI Messages
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need for effective annotation and standardization to identify images or videos created or modified by generative artificial intelligence, as current methods struggle to distinguish AI-generated content from original content.
Innovation Solution
The proposed solution involves using Supplementary Enhancement Information (SEI) messages in video bitstreams to mark the involvement of generative AI in the creation or modification of images or videos. This is achieved by setting a specific value for a text description purpose parameter and including AI marking information within the SEI message.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If generative AI is used to create or modify images or videos, then content creation capability is improved, but ability to distinguish AI-generated content from original content deteriorates
Solution Approach 1:
The patent introduces an intermediary marking mechanism (SEI messages with specific parameters) that mediates between the AI generation process and the content consumption process. This marking layer allows AI-generated content to be identified without affecting the visual quality or creating obvious artifacts, resolving the contradiction between high-quality AI generation and detectability.
Solution Approach 2:
The patent changes the parameter space of video bitstreams by introducing new SEI message parameters (text description purpose parameter, text description information string) that carry AI generation metadata. This allows differentiation of AI-generated content through parameter modification rather than visual alteration, maintaining content quality while enabling detection.
2Loss of information
If AI marking information is added to video bitstreams, then content identification capability is improved, but bitstream complexity increases
Solution Approach 1:
The patent makes the SEI message structure universal by using existing video coding frameworks (H.266/VVC) to carry multiple functions: compression, AI generation marking, and content identification. This multi-functionality avoids creating separate complex marking systems, resolving the contradiction between information preservation and system complexity.
Solution Approach 2:
The patent nests the AI marking information within the existing SEI message structure of the video bitstream. The marking data is embedded inside the SEI message container, which itself is part of the larger video bitstream structure. This nested arrangement minimizes additional complexity by reusing existing container structures rather than creating top-level complex systems.
3Quantity of substance
If compression is applied to reduce bandwidth and storage requirements, then data efficiency is improved, but accuracy of AI content detection may deteriorate
Solution Approach 1:
The patent applies preliminary action by embedding AI marking information into the compressed bitstream during the encoding process, before the content is transmitted or stored. This ensures that the detection metadata is preserved through compression without requiring additional processing steps later, maintaining detection accuracy while achieving data efficiency through standard compression.
Solution Approach 2:
The patent uses SEI messages as a copying mechanism to replicate AI generation metadata through the compression process. The marking information is copied into the bitstream structure where it survives compression transformations, allowing detection accuracy to be maintained independently of the compression ratio applied to the visual content.
Data Source
AI summary
A method and apparatus comprising computer code for video processing, the method including setting a first value of a text description purpose parameter in a bitstream, the first value of the text description purpose parameter indicating a type of information included in a text description information string in the bitstream; setting artificial intelligence (AI) marking information associated with one or more pictures in the bitstream as the text description information string when the first value indicates that the type of information included in the text description information string comprises marking information associated with one or more artificial intelligence processes used; signaling the text description purpose parameter in the bitstream; and signaling the text description information string.