Selective Face De-Anonymization for Target Interaction Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition solutions face missed detection due to algorithm configuration and training dataset limitations, leading to privacy concerns as they indiscriminately anonymize data, hiding interactions of a target with other subjects.
Innovation Solution
A method and apparatus that retrieve historical images to identify interactions between a target and subjects, canceling anonymization for the target and identified subjects by comparing input images and historical data, using background subtraction and face recognition to de-anonymize relevant areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If anonymization is applied to all detected faces for data privacy protection, then privacy protection is improved, but detection accuracy and interaction identification deteriorate
Solution Approach 1:
The patent applies different quality levels of anonymization to different regions. Faces that have interacted with the target are de-anonymized (high quality), while other faces remain anonymized (low quality). This local differentiation resolves the contradiction by maintaining privacy protection for non-interacting individuals while ensuring accurate detection and interaction identification for relevant targets.
2Productivity
If face detection algorithms are configured with standard parameters, then processing efficiency is improved, but detection accuracy deteriorates due to missed detections
Solution Approach 1:
The patent performs preliminary face detection using standard efficient algorithms, then applies a secondary verification process using historical image comparison and interaction detection. This preliminary action maintains processing efficiency while the subsequent verification step compensates for missed detections, improving overall detection accuracy without sacrificing speed.
3Object-affected harmful factors
If anonymization is applied to all subjects in the image, then data privacy is improved, but information utility for contact tracing deteriorates
Solution Approach 1:
The patent extracts and isolates the specific information needed for contact tracing - the interaction history between the target and other subjects. By separating this critical information from the general anonymized image data, the system maintains privacy protection for the majority of subjects while preserving essential interaction information for contact tracing purposes.
4Measurement precision
If historical images are retrieved and compared to identify interactions, then interaction detection accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent applies partial action by retrieving and comparing only the necessary historical images related to the target subject, rather than processing all historical data. This selective approach maintains interaction detection accuracy for relevant interactions while significantly reducing processing time and computational resource consumption compared to comprehensive analysis.
Data Source
AI summary
In an aspect, there is provided method for cancelling anonymization for an area including a target in an input image, the input image including a plurality of identified subjects, the method comprising: retrieving historical images in which the target has appeared in; identifying an area by comparing the input image and the retrieved historical images to determine if the target has interacted with at least one of the identified subjects; and cancelling anonymization for the area including the target.