Biometric Group Identification Using Machine Learning Templates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial matching technologies in biometric identification face challenges such as low match rates due to varying face angles, security risks from storing biometric data, and the potential for reverse engineering of biometric templates, particularly in high-security environments like intelligence or law enforcement settings.
Innovation Solution
A device and method utilizing machine-learning developed biometric templates with adjustable match scores and two-stage matching processes, where a high match threshold is applied to an authorized group and a lower threshold to a watch list, ensuring high match rates without storing photographs and making it difficult to reconstruct biometrics from the machine learning templates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional biometric templates are stored in a database, then match rates can be achieved, but the system becomes vulnerable to reverse engineering and database compromise
Solution Approach 1:
The patent uses machine learning models to learn the distribution and characteristics of biometric data without storing actual biometric templates or photographs. The system creates a statistical copy or representation of the biometric data distribution that enables matching while preventing reconstruction of original biometric information.
Solution Approach 2:
The machine learning model acts as an intermediary between the stored biometric data and the matching process. Instead of directly comparing stored templates with new inputs, the system uses the trained model to evaluate whether new biometric data belongs to the same group, thereby preventing direct access and reverse engineering of original templates.
2Reliability
If a high match threshold is used for group identification, then false acceptances are reduced, but match rates decrease
Solution Approach 1:
The patent changes the fundamental parameter being evaluated from template similarity scores to probabilistic group membership predictions. The machine learning model outputs a probability or confidence score that an individual belongs to the group, allowing for optimized threshold selection that balances false acceptances and match rates based on the specific security requirements.
3Measurement precision
If numerous biometric templates are stored for each individual, then match accuracy improves, but security risks and database compromise value increase
Solution Approach 1:
The system creates a machine learning model that captures the essential characteristics and variations of multiple biometric templates without storing the templates themselves. The model learns from numerous examples during training, achieving high match accuracy, while the stored representation remains incomprehensible and useless for reconstructing original biometric data.
Data Source
AI summary
A device and method for achieving a high probability of match for identifying individuals in a particular group, is provided. The device and method consists of two stages. The first stage compares a biometric template against personnel in an authorized group with a high probability of match standard. In the second stage, in response to no match being made in the authorization group, a search would be conducted against a second group, such as a watch list, with the same or lower probability of match rate.