Perceptual-Map Image Perturbation for Generative AI Learning Prevention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image generation technologies using generative AI models face challenges in protecting copyrighted data from unauthorized use and imitation during the learning process.
Innovation Solution
A learning prevention method is implemented using a perceptual map to represent perceptual sensitivity, inserting subtle perturbations into original images based on this map to generate a result image that prevents unauthorized learning by generative AI models, while maintaining image quality through perceptual constraint pools.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If perturbation is inserted into the original image to prevent learning by generative AI, then copyright protection is improved, but image quality deteriorates
Solution Approach 1:
The patent applies local quality by inserting perturbations selectively in specific regions of the image based on a perceptual map. The perceptual map identifies regions where perturbations are less noticeable to human observers, allowing stronger perturbation insertion in those areas while maintaining image quality in perceptually sensitive regions. This resolves the contradiction by making the protection effect local to non-critical areas while preserving overall image quality.
Solution Approach 2:
The patent changes the parameter of perturbation strength based on the perceptual map values. Different regions of the image receive different perturbation intensities according to their perceptual sensitivity. This parameter adaptation allows the system to maximize copyright protection effectiveness while minimizing visible degradation, thereby resolving the contradiction between protection strength and image quality.
2Reliability
If strong perturbation is inserted to effectively block learning, then learning prevention effectiveness is improved, but perceptual noticeability worsens
Solution Approach 1:
The patent uses local quality by varying perturbation strength across different image regions based on the perceptual map. Regions with lower perceptual sensitivity receive stronger perturbations for effective learning prevention, while regions with high perceptual sensitivity receive weaker or no perturbations to avoid noticeability. This spatially differentiated approach resolves the contradiction between effectiveness and noticeability.
Solution Approach 2:
The perceptual map serves as an intermediary that guides perturbation insertion. It acts as a mediator between the goal of strong perturbation (for effectiveness) and the constraint of minimal noticeability. By using the perceptual map as an intermediary, the system can intelligently distribute perturbation strength, achieving both effectiveness and low noticeability simultaneously.
Data Source
AI summary
Disclosed are a learning prevention methods, a computer device configured to perform the learning prevention method, and a recording medium storing instructions to perform the learning protection methods may be provided. The learning prevention method may include generating a perceptual map for an original image, the perceptual map representing perceptual sensitivity to perturbation for the original image, and inserting the perturbation into the original image based on the perceptual map to generate a result image for learning prevention.