Face Vector Storage and Retrieval via Latent Space Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The effectiveness of facial recognition technology is hindered by the volume of facial data collected, and differing security and privacy requirements across organizations complicate data retrieval and security in shared environments, leading to issues with data access and protection.
Innovation Solution
The technology compacts face-related data by generating representative vectors in a latent space, allowing for efficient grouping and retrieval of face vectors, and uses homomorphic encryption to secure data access, enabling organizations to control access to their face-related data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If face-related data is stored in full detail, then data completeness is improved, but data retrieval speed deteriorates due to the large volume of data
Solution Approach 1:
The patent extracts essential features from full face images to create face vectors that capture the most important identifying characteristics. This extraction process removes redundant information while preserving the core data needed for recognition, thereby maintaining data completeness while reducing the volume that slows retrieval operations.
Solution Approach 2:
The patent creates simplified copies of face data in the form of face vectors and representative vectors. These vectors are mathematical representations that replicate the essential identifying features of faces without requiring the full image data, enabling fast retrieval while preserving recognition accuracy.
2Productivity
If face vectors from different organizations are stored together, then data retrieval efficiency is improved, but data security and privacy protection deteriorate
Solution Approach 1:
The patent segments the vector collection system into organization-specific collections, where each organization's face vectors are stored in separate, controlled access collections. This segmentation allows efficient retrieval within each organization while maintaining security boundaries that protect privacy and prevent unauthorized access across organizations.
Solution Approach 2:
The patent introduces encrypted representative vectors as intermediaries between search queries and actual face vectors. These encrypted vectors act as a mediator that enables efficient similarity search without exposing the underlying sensitive data, allowing retrieval efficiency while maintaining security through the intermediary encryption layer.
3Measurement precision
If detailed face data is stored for accurate recognition, then recognition accuracy is improved, but storage complexity and access control difficulty increase
Solution Approach 1:
The patent transforms face data from image space to vector space, changing the parameters from pixel values to mathematical vectors that capture essential features. This parameter transformation maintains recognition accuracy by preserving the most discriminative features while reducing storage complexity through more compact vector representations.
Solution Approach 2:
The patent moves face data from the traditional two-dimensional image space into a high-dimensional vector space where facial features are represented as points. This dimensional transformation enables more efficient storage and indexing while maintaining recognition accuracy through the geometric relationships in the vector space.
Data Source
AI summary
A method includes generating a first representative vector based on a first vectors, wherein the first representative vector is associated with the first vectors in a collection of representative vectors, and the first vectors comprises a set of vector values within a latent space. The method further includes generating a second representative vector based on a second vectors, wherein the second representative vector is associated with the second vectors in the collection of representative vectors. The method further includes determining a latent space distance based on the first and second vectors. The method further includes determining whether the latent space distance satisfies a threshold. In response to a determination that the latent space distance satisfies the threshold, the method further includes associating a combined representative vector with the first vectors and the second vectors and removing the first and second representative vectors from the collection of representative vectors.