Biometric Authentication Architecture Using Difference Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Modern biometric authentication systems based on artificial intelligence models are vulnerable to security attacks due to the storage of learned structures, which can be exploited by malicious users to produce synthetic inputs that bypass authentication.
Innovation Solution
A novel approach involving a blended model with a classifier producing output vectors in a vector space, partitioned fixed points, and storage of difference vectors, utilizing expanders to enhance security and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If learned structures are stored in non-volatile memory for authentication, then authentication accuracy is improved, but security vulnerability increases
Solution Approach 1:
The patent extracts the harmful stored learned structures from the system by introducing a fuzzy extractor that generates cryptographic keys without storing the actual biometric templates or learned models. The extractor processes biometric data through a reference value and randomness to produce keys that can be stored securely without exposing the underlying biometric information or model structures.
Solution Approach 2:
The patent introduces a fuzzy extractor as an intermediary component between the biometric authentication system and the storage mechanism. This intermediary transforms the raw biometric data and learned structures into cryptographic keys that can be stored in non-volatile memory without revealing the original biometric information or model structures, thus mediating between authentication needs and security requirements.
2Productivity
If similarity-score algorithms are used for biometric authentication, then authentication performance is improved, but vulnerability to synthetic input attacks increases
Solution Approach 1:
The patent converts the vulnerability to synthetic input attacks into a benefit by using the fuzzy extractor to generate cryptographic keys that are inherently resistant to such attacks. The extractor's use of reference values and randomness ensures that even if an attacker has access to stored keys or model information, they cannot generate synthetic inputs that would succeed in authentication, as the keys are derived through a secure cryptographic process.
3Reliability
If biometric data and model information are stored for authentication, then authentication capability is maintained, but information leakage risk increases
Solution Approach 1:
The patent extracts and removes the sensitive biometric data and model information from the storage system by using a fuzzy extractor that generates cryptographic keys without retaining the original data. The extractor processes biometric inputs through a reference value and randomness, producing keys that can be stored without exposing the underlying biometric templates or learned models, thus eliminating information leakage risks while maintaining authentication capability.
Solution Approach 2:
The patent introduces a fuzzy extractor as an intermediary that stands between the biometric data and the storage mechanism. This intermediary transforms sensitive information into cryptographic keys that can be stored securely, preventing direct storage of biometric data or model information and thereby reducing information leakage risks while preserving authentication functionality.
Data Source
AI summary
Computer-implemented methods, systems and computer-readable media for building and using an artificial intelligence model for secure biometric authentication. Utilizing difference vectors, the model securely relates output vectors generated from noisy biometric data of a plurality of enrolled users to pre-defined fixed points in a vector space.