Neural Video Fingerprinting for Blockchain Fake-Video Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies are inadequate in detecting and preventing the alteration of video images to defame individuals, as they lack effective methods to authenticate the genuineness of videos, particularly those manipulated using deep learning algorithms.
Innovation Solution
A system that utilizes neural networks for image and video analysis to detect alterations by identifying spatial and frequency domain irregularities, combined with blockchain technology to authenticate videos through digital fingerprinting and embedding hashes in frames.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If deep learning algorithms are used to alter video images, then the ability to create entertaining content is improved, but the ability to defame individuals and create fake videos worsens
Solution Approach 1:
The patent applies preliminary action by embedding authentication data (hash values) into the original video frames before distribution. This proactive measure allows the video to be verified as authentic later, preventing the spread of manipulated content while still allowing deep learning-based content creation to proceed.
Solution Approach 2:
The patent introduces an intermediary authentication mechanism using hash values and blockchain technology. This intermediary layer verifies the authenticity of video content without interfering with the creative deep learning processes, enabling both content creation versatility and protection against misuse.
2Reliability
If video authentication methods are implemented, then the reliability of video content is improved, but the device complexity and processing requirements worsen
Solution Approach 1:
The patent segments the authentication process into distinct components: generating hash values from video frames, embedding these hashes into the video data, and verifying authenticity by comparing extracted hashes with expected values. This segmentation makes the complex authentication process more manageable and implementable.
Solution Approach 2:
The patent extracts essential authentication information (hash values) from the video content and embeds them in a separate, verifiable manner. This extraction approach allows authentication to be performed without reprocessing the entire video, reducing computational complexity while maintaining reliability.
3Measurement precision
If hash values are embedded in every video frame, then the detection precision of fake videos is improved, but the loss of time and processing energy worsens
Solution Approach 1:
The patent applies partial action by embedding hash values in selected video frames rather than every single frame. This selective approach maintains sufficient detection precision while significantly reducing the time and processing energy required for authentication compared to embedding hashes in all frames.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Detection of whether a video is a fake video (212) derived from an original video (200) and altered is undertaken using a block chain that either forbids (912) adding to the block chain copies of original videos that have been altered or indicating in the block chain that an altered video has been altered. Image fingerprinting techniques are described for determining (908, 1002) whether video sought to be added to block chain has been altered.