Variational Autoencoder Compact Representations for Biometric Enrollment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Biometric identification systems face challenges in maintaining user privacy and efficiency due to the need to store sensitive input data, which can be compromised and require re-enrollment upon updates to the system.
Innovation Solution
A compact representation module using a variational autoencoder and embedding model is trained to generate a compact representation of biometric data, eliminating the need to store raw images, while allowing for updates without re-enrollment by using a loss function that minimizes data size and maintains feature richness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If raw biometric input data is stored for identification, then identification accuracy is maintained, but user privacy is compromised and storage requirements increase
Solution Approach 1:
The patent extracts only the essential features from raw biometric input data to create feature vectors, discarding the original sensitive images. This extraction process maintains identification accuracy while eliminating privacy risks associated with storing raw biometric data.
Solution Approach 2:
The patent creates compressed representations (feature vectors) that copy only the essential information needed for identification, rather than storing complete raw images. These feature vectors serve as sufficient copies for the identification function while occupying minimal storage space.
2Adaptability or versatility
If raw biometric images are stored for system updates, then model retraining is enabled, but storage requirements and computational overhead increase
Solution Approach 1:
The patent extracts feature vectors from raw images during enrollment, storing only these compact features. When system updates are needed, the extracted features are used for retraining models, eliminating the need to store and process large volumes of original images while maintaining adaptability.
3Productivity
If complete input data is retained for analysis, then comprehensive processing is possible, but computational requirements and processing time increase
Solution Approach 1:
The patent extracts essential features into compact vectors before storage and processing. This extraction dramatically reduces the data volume that needs to be processed during identification and analysis operations, thereby improving processing efficiency and reducing computational energy consumption.
Data Source
AI summary
During enrollment to a biometric identification system an image of at least a portion of a user is acquired and processed to determine a compact representation (CR). The CR may be stored for later comparison to identify the user, to train embedding models, and so forth. The image is not stored. A machine learning system that comprises a variational autoencoder is trained to produce the CR with a loss function that includes a distortion loss, an embedding distance loss, and in some implementations a bitrate loss. The trained encoder is used to determine the CR from an input image, while the decoder is not stored. The CR contains sufficient information to be used as training data for embedding models, instead of acquired images. Training may be more computationally efficient using the CR. Recognition comparisons may be more efficiently performed using the CR, compared to query images.