Face Recognition With Distinctiveness Mapping for Deidentification Resistance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition systems are vulnerable to adversarial attacks through deidentification methods that manipulate specific facial features, reducing their accuracy and posing privacy concerns.
Innovation Solution
A machine learning-based approach that identifies and masks or modifies highly distinctive facial features, using a distinctiveness map to train a model to recognize faces despite deidentification attempts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deidentification methods manipulate specific facial features to protect privacy, then privacy protection is improved, but facial recognition accuracy deteriorates
Solution Approach 1:
The patent segments the face into multiple regions and identifies distinctive regions that are most informative for recognition. By focusing on these specific segments rather than the entire face, the system can protect privacy in non-distinctive regions while maintaining recognition accuracy through distinctive regions.
Solution Approach 2:
The patent applies different processing qualities to different facial regions. Distinctive regions receive higher processing quality for recognition, while non-distinctive regions can be more heavily modified for privacy protection. This local differentiation resolves the contradiction by optimizing both privacy and recognition in their respective zones.
2Reliability
If deidentification methods condense changes at regions in focus of FR systems, then privacy protection is improved, but facial recognition accuracy deteriorates
Solution Approach 1:
The patent performs preliminary analysis to identify distinctive regions before applying deidentification modifications. By pre-mapping which regions are critical for recognition, the system can strategically place privacy-protecting modifications in non-critical regions, achieving effective deidentification without compromising recognition accuracy.
3Adaptability or versatility
If adversarial attacks modify input images to cause FR classifiers to fail, then security testing is improved, but system reliability deteriorates
Solution Approach 1:
The patent implements preliminary anti-action by training the FR system with adversarial examples during the training phase. The distinctive region mapping and modified training images prepare the system in advance to resist adversarial attacks, maintaining reliability while enabling security testing through the distinctive region analysis capability.
Data Source
AI summary
A system and a method of performing face recognition may include: receiving a first facial image, depicting a first face, and a second facial image depicting a second face; applying an ML model on the first image, to produce a first representation vector, and applying the ML model on the second image to produce a second representation vector; comparing the first representation vector and the second representation vector; and associating the first face with the second face based on the comparison, where the ML model is trained to produce the representation vectors from the facial images, based on regions in the facial images that correspond to distinctiveness scores that are beneath a distinctiveness threshold.