Whole-Image Embeddings for Privacy-Preserving Identity Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition systems violate privacy by processing images without consent, leading to potential violations of biometric regulations and inefficiencies in handling multiple individuals.
Innovation Solution
A system that processes entire images using whole-image embedding representations (WIER) trained on general image understanding tasks, preserving privacy by not isolating individuals, and using a combination of image and text generation models to generate a single embedding vector for multiple people, allowing for efficient identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial recognition systems process images to identify individuals, then identification accuracy is improved, but privacy protection deteriorates
Solution Approach 1:
The patent extracts only the necessary identification features from images while deliberately excluding sensitive personal information. The system processes images to generate embedding representations that capture identity-relevant patterns without isolating or storing identifiable facial features, thus achieving identification while protecting privacy.
Solution Approach 2:
The patent introduces embedding representations as an intermediary between the original image and the identification result. These embeddings serve as a mediator that preserves identification capability while removing direct links to identifiable individuals, allowing the system to identify without exposing private information.
2Measurement precision
If traditional facial recognition systems isolate and process individual faces, then identification precision is improved, but system complexity increases
Solution Approach 1:
The patent merges the processing of multiple individuals in an image into a single unified embedding representation. Instead of isolating and processing each face separately, the system processes the entire image as a whole, combining all individual representations into one collective embedding that preserves identification information while simplifying the processing pipeline.
Solution Approach 2:
The patent creates a universal embedding representation that serves multiple functions simultaneously: it enables identification of multiple individuals, maintains privacy protection, and works across different image compositions. This single embedding structure replaces the need for separate processing pipelines for each individual.
3Object-affected harmful factors
If whole-image embedding representations are used to preserve privacy, then privacy protection is improved, but identification reliability may deteriorate
Solution Approach 1:
The patent transforms the representation parameters from detailed facial features to aggregated embedding vectors. By changing the parameter space from high-resolution facial details to compressed embedding representations, the system maintains identification reliability through the preserved structural relationships while improving privacy protection through information aggregation.
Data Source
AI summary
The system receives data indicating an individual and processes the data to isolate the individual and to enhance data quality. The system extracts a first multiplicity of key features of the data, which tend to uniquely identify the individual. The system compares, using artificial intelligence, the first multiplicity of key features associated with the data to a second multiplicity of key features associated with a user to determine whether the data indicates the user. Upon determining that the data indicates the user, the system retrieves from a datastore a rule associated with the second multiplicity of key features and determines whether the rule permits use of the data indicating the individual. Upon determining that the rule associated with the second multiplicity of key features does not permit use of the data indicating the individual, the system sends an indication that the rule does not permit the use.