Biometric Representation Storage for Privacy-Safe Model Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional biometric identification systems face challenges in safeguarding privacy and security of user data, as storing input data as a 'gallery' raises concerns about impermissible access, and updating embedding models requires users to re-enroll, leading to inconvenience and increased costs.
Innovation Solution
The system stores representation data instead of input data, using a trained representation module to process image data and generate representation data that cannot be computationally reconstructed, allowing updates without re-enrollment and reducing storage and computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If input data is stored as a gallery for biometric identification, then the system can perform recognition and identity assertion, but privacy and security concerns arise due to potential impermissible access
Solution Approach 1:
The patent extracts only the essential features from input biometric data to create compact representations (embeddings), storing only these extracted features rather than the complete original data. This allows the system to maintain identification functionality while eliminating the privacy and security risks associated with storing comprehensive biometric galleries.
Solution Approach 2:
The patent creates simplified copies of biometric data in the form of numerical embeddings or feature vectors that capture the essential characteristics needed for identification. These compact representations serve as functional substitutes for the original biometric data, enabling recognition while reducing the harm potential of stored information.
2Measurement precision
If embedding models are updated to improve identification performance, then recognition accuracy increases, but users must re-enroll which causes inconvenience and increases costs
Solution Approach 1:
The patent performs preliminary feature extraction during the initial enrollment process, creating compact representations that are model-agnostic. These pre-extracted features can be processed by different embedding models without requiring users to re-enroll, allowing model updates while maintaining user convenience.
Solution Approach 2:
The patent separates the feature extraction process from the embedding generation process. The feature extraction creates intermediate representations that can be independently processed by different embedding models, allowing model updates without requiring users to repeat the entire enrollment process.
3Adaptability or versatility
If complete input data is stored for future model updates, then training new embedding models is possible, but storage requirements and computational costs increase
Solution Approach 1:
The patent extracts only the essential feature representations needed for model training and storage, discarding redundant information. This extraction process creates compact data structures that enable future model updates while significantly reducing storage requirements compared to storing complete input datasets.
Solution Approach 2:
Instead of storing complete input data and processing it when needed, the patent inverts the approach by pre-processing and storing only the essential extracted features. This reversal allows model updates with minimal storage requirements, as the compact feature representations can be efficiently reused across different embedding models.
Data Source
AI summary
During operation a biometric identification system acquires an image of at least a portion of a user that is processed by a representation model to determine representation data. The representation data may be securely stored for later use. The image is not stored. The stored representation data is smaller than equivalent image data. The representation model may comprise a subset of a first recognition model that has been trained using image training data. The image training data may be processed by the trained representation model to determine training representation data. The representation data contains sufficient information to be used as training data for recognition models, instead of acquired images. Training of recognition models using representation data is more computationally efficient compared to training with image data. Recognition comparisons using query representation data are more computationally efficient compared to query image data.