Whole-Image Identity Matching Under Partial Privacy Approval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial recognition systems violate privacy by processing and isolating individual faces without consent, leading to potential legal issues and reduced accuracy when dealing with multiple individuals in a scene.
Innovation Solution
A system that uses whole-image representations (WIER) to identify multiple individuals in an image without isolating them, using machine learning components trained on general image understanding tasks, combining image and text embeddings to generate a single representation for privacy-preserving identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial recognition systems process and isolate individual faces, then identification accuracy can be improved, but user privacy is violated and legal compliance deteriorates
Solution Approach 1:
The patent segments the identification task by separating face isolation processing from whole-image analysis. Instead of extracting and processing individual face regions (traditional approach), the system divides the image into multiple candidate regions and processes them independently through the same neural network, allowing privacy-preserving identification by treating all regions equally without prioritizing specific individuals
Solution Approach 2:
The patent applies a universal neural network model that can process both whole images and candidate regions through the same architecture. The system uses a single trained model for multiple purposes: generating whole-image embeddings for privacy-preserving identification and processing candidate regions for traditional identification, eliminating the need for separate face-specific processing pipelines
2Reliability
If facial recognition systems isolate individual faces for processing, then identification performance can be improved, but system complexity increases due to multiple processing pipelines
Solution Approach 1:
The patent employs a universal neural network architecture that handles both whole-image processing and candidate region processing through the same model. This single model serves multiple functions: generating embeddings for privacy-preserving identification and processing individual candidate regions, thereby reducing system complexity by eliminating the need for separate face-specific processing pipelines while maintaining identification performance
Solution Approach 2:
The patent merges the traditionally separate face isolation pipeline and whole-image analysis pipeline into a unified system. Both processing paths use the same neural network model and can be combined through embedding comparison, simplifying the overall system architecture while maintaining the benefits of both approaches
3Measurement precision
If facial recognition systems use face-specific processing, then identification accuracy is maintained, but adaptability to different privacy requirements deteriorates
Solution Approach 1:
The patent implements a dynamic processing approach where the system can adapt between privacy-preserving mode (using whole-image embeddings) and traditional mode (using candidate region embeddings) based on privacy requirements. The same neural network model supports both modes, allowing flexible adaptation to different legal and privacy contexts without sacrificing recognition accuracy in either mode
Data Source
AI summary
The system obtains a first image including a first individual and an indication that the first individual can be identified in the first image. Upon obtaining the indication, the system processes the first image using a first component and obtains a first identity embedding. The system obtains a second image including a second individual and other objects without obtaining an indication that the second individual can be identified. Upon obtaining the indication, the system provides the second image to a second component and obtains a first WIER representing the second image without isolating the second individual. The system provides the first identity embedding to a third component configured to convert the first identity embedding into a second WIER. The system determines whether the second image includes the first individual by determining whether the first and the second WIER match.