Transformer-Based Embedding Space Conversion for Biometric Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional biometric identification systems face challenges in updating their algorithms without re-enrolling users, as they need to retain gallery data to maintain system accuracy and privacy, which is costly and inconvenient.
Innovation Solution
The use of transformer networks to convert first embeddings into second embeddings without storing the gallery data, allowing for system updates and user migrations while maintaining privacy and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If gallery data is stored to maintain system accuracy and enable algorithm updates, then system reliability and adaptability are improved, but data privacy protection and storage cost worsen
Solution Approach 1:
The patent extracts only the essential embedding representations from the gallery data while discarding the actual biometric images and raw data. By storing only the processed embedding vectors rather than the original data, the system maintains the ability to perform accurate identification while eliminating the privacy risks associated with storing sensitive biometric information.
Solution Approach 2:
The patent creates a transformed copy of the gallery data in the form of embedding representations. These embeddings serve as a compressed, processed copy that retains the essential identification information needed for system accuracy while being much safer to store compared to the original biometric data.
2Adaptability or versatility
If gallery data is retained to enable algorithm updates without re-enrollment, then adaptability is improved, but storage cost and data retention time worsen
Solution Approach 1:
The patent extracts only the necessary embedding information from the full gallery data, creating a reduced representation that is sufficient for algorithm updates. This extracted data can be transformed between different embedding spaces without requiring the full original gallery, significantly reducing storage requirements while maintaining update capability.
Solution Approach 2:
The patent changes the parameter representation by transforming data between different embedding spaces. By maintaining data in multiple embedding representations, the system can adapt to different algorithms and models without storing all possible versions of the original data, reducing storage cost while enabling adaptability.
3Measurement precision
If gallery data is stored to maintain identification accuracy, then measurement precision is improved, but data retention duration worsens
Solution Approach 1:
The patent extracts only the essential embedding representations that are needed for identification accuracy, separating this functional requirement from the need to retain original biometric data. The extracted embeddings can be maintained indefinitely for accuracy while the original sensitive data can be deleted, effectively reducing retention time of sensitive information.
Data Source
AI summary
An image of at least a portion of a user during enrollment to a biometric identification system is acquired and processed with a first model to determine a first embedding that is representative of features in that image in a first embedding space. The first embedding may be stored for later comparison to identify the user, while the image is not stored. A second model that uses a second embedding space may be later developed. A transformer is trained to accept as input an embedding from the first model and produce as output an embedding consistent with the second embedding space. The previously stored first embedding may be converted to a second embedding in a second embedding space using the transformer. As a result, new embedding models may be implemented without requiring storage of user images for later reprocessing with the new models or requiring re-enrollment by users.