AI Face Masking for Privacy-Preserving Subject Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing surveillance camera technologies anonymize faces, impairing the subject recognition function, and there is a need for a method that anonymizes subjects without compromising this function.
Innovation Solution
An image processing device and method that utilize artificial intelligence to convert region images of subjects into feature images, generating a mask image by combining the original and feature images, preserving subject recognition capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If face anonymization is performed by existing techniques, then personal information protection is improved, but subject recognition function is impaired
Solution Approach 1:
The patent segments the image processing into distinct functional components: face detection module, feature extraction module (encoding to latent space), and image reconstruction module (decoding back to image space). This segmentation allows independent optimization of each module - the encoding module protects privacy by transforming facial features, while the decoding module preserves subject recognition capabilities through learned feature mappings.
Solution Approach 2:
The patent introduces an intermediary latent feature representation that acts as a mediator between the original image and the anonymized output. This latent space encoding serves as an intermediary transformation that preserves essential subject characteristics while removing identifiable facial features, enabling both privacy protection and subject recognition to coexist.
2Object-affected harmful factors
If traditional anonymization methods are used, then privacy protection is improved, but identification capability is worsened
Solution Approach 1:
The patent changes the parameter representation of facial features by transforming them from pixel-space to latent feature-space through neural network encoding. This parameter transformation preserves the essential characteristics needed for identification while altering the specific facial appearance, thereby protecting privacy without compromising identification capability.
Solution Approach 2:
The patent creates a composite representation by combining multiple feature dimensions in the latent space - some dimensions preserve subject-specific characteristics for identification, while others are transformed or obscured for privacy protection. This composite feature representation enables simultaneous achievement of both goals.
Data Source
AI summary
An image processing device includes circuitry that recognizes and extracts an object included in a captured image, and that converts, using an artificial intelligence (AI) model, a region image including the object to generate a feature image. The circuitry generates a mask image by combining the captured image with the feature image.