AI Virtual Face Generation System Using Autoencoder Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face synthesis technologies using artificial intelligence are risky due to potential abuse and issues with portrait rights, as they often generate recognizable faces of real individuals, making them unsuitable for widespread use in content creation.
Innovation Solution
A method and device utilizing autoencoder technology in deep learning to generate virtual faces by combining face source data and background features, allowing for the creation of unique virtual persons that do not exist in the world, thereby avoiding recognizable individuals and minimizing portrait rights issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If face synthesis technology is used to create realistic faces, then content creation capability is improved, but the risk of abuse and portrait rights violations increases
Solution Approach 1:
The patent introduces an intermediary verification system that acts as a mediator between face synthesis technology and its application. This system includes a verification module that checks generated faces against a database of real person faces, and a blocking module that prevents generation of faces that match real individuals. This intermediary mechanism enables content creation while filtering out harmful outputs that would violate portrait rights or enable abuse.
Solution Approach 2:
The patent implements a feedback mechanism where the verification module continuously monitors generated faces and provides feedback to the blocking module. When a generated face is detected to match a real person, the system learns from this feedback and adjusts its blocking parameters. This feedback loop enables the system to improve its ability to prevent abuse while maintaining content creation capabilities.
2Manufacturing precision
If deep learning models are trained on existing face data, then face generation quality is improved, but the likelihood of generating recognizable real persons increases
Solution Approach 1:
The patent replaces the traditional mechanical approach of directly training deep learning models on face databases with a substituted system that includes verification and blocking modules. Instead of relying solely on the training data composition to prevent recognizability, the system uses an additional verification layer that checks generated faces against real person databases and blocks matches. This substitution enables high-quality generation while maintaining control over recognizability.
3Productivity
If face synthesis technology is made widely accessible, then content creation productivity is improved, but the potential for harmful use increases
Solution Approach 1:
The patent applies preliminary action by implementing verification and blocking mechanisms before face synthesis output is released. The system proactively checks generated faces against databases of real persons and blocks potentially harmful outputs before they can be used. This preliminary verification enables widespread accessibility for legitimate content creation while preventing harmful uses in advance.
Data Source
AI summary
Provided are a method and a device for generating a virtual face, the method including: performing, by a device for generating the virtual face, comparison and learning on an inferred face and an existent face through deep learning after receiving a plurality of pieces of face source data and at least one piece of face background data; and generating, after receiving one piece of face background data, virtual face data by combining the face inferred from the plurality of pieces of face source data with a feature of the one piece of face background data through a model generated by comparison and learning.