Universal Discriminator for De-identification Degree Scoring

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current deep-learning based de-identification technologies face challenges in determining the degree of de-identification for various types of data, leading to potential personal information leakage due to incomplete or unsuccessful de-identification processes.

Innovation Solution

A method is developed to train a universal discriminator capable of generating discriminant scores for each image based on its de-identification degree, using both successfully and unsuccessfully de-identified images, and an obfuscation network is trained using this discriminator to enhance de-identification performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If deep-learning based de-identification technologies are used, then de-identification processing can be automated, but the ability to accurately determine the degree of de-identification is insufficient leading to potential personal information leakage

Engineering Contradiction:

Improveautomation of de-identification processingVSAvoidaccuracy of determining de-identification degree

Core Design Contradiction:

Extent of automationVSMeasurement precision

Solution Approach 1:

A discriminator network is introduced as an intermediary component to evaluate the degree of de-identification. The discriminator receives obfuscated images as input and outputs a determination result indicating whether personal information is sufficiently concealed, thereby providing automated yet accurate measurement of de-identification effectiveness

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements feedback mechanisms where the discriminator's evaluation results are used to guide the optimization of the obfuscation network. The loss function incorporates the discriminator's output to adjust obfuscation parameters, creating a closed-loop system that continuously improves de-identification accuracy while maintaining automation

Inventive Principle:

Principle #23Feedback

2Productivity

If de-identification processing is applied to vast amounts of big data, then data utility is maintained, but the risk of incomplete de-identification and personal information leakage increases

Engineering Contradiction:

Improveprocessing volume of de-identificationVSAvoidconsistency of de-identification effectiveness

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system employs self-service mechanisms where the discriminator automatically evaluates each processed image and provides real-time feedback on de-identification quality. This enables the system to monitor and ensure consistent effectiveness across large volumes of processed data without requiring manual verification for each item

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

Manual inspection methods are replaced with automated deep-learning-based evaluation. The discriminator network processes images through neural network layers to automatically determine de-identification degree, replacing mechanical human review processes and enabling scalable processing while maintaining reliability through consistent algorithmic evaluation

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

3Adaptability or versatility

If multiple de-identification methods are applied to different data types, then coverage of personal information is improved, but the complexity of determining de-identification degree increases

Engineering Contradiction:

Improvecoverage of different data typesVSAvoidcomplexity of evaluation system

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The discriminator is designed as a universal evaluation system that can assess de-identification effectiveness across multiple data types including images, text, and audio. The network architecture uses shared layers for common feature extraction followed by type-specific evaluation layers, enabling multi-functional capability while managing complexity through modular design

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS11308359B1Methods for training universal discriminator capable of determining degrees of de-identification for images and obfuscation network capable of obfuscating images and training devices using the same

Publication Date: 2022.04.19 DEEPING SOURCE INC

AI summary

Methods for training a universal discriminator and an obfuscation network are provided. The methods include steps of: generating an obfuscated image by obfuscating an original image through the obfuscation network, characteristic information by applying learning operation to the obfuscated image through a surrogate network, a first discriminant score determining a degree of de-identification for the obfuscated image through the universal discriminator, and a second discriminant score determining whether the obfuscated image is real or fake through a regular discriminator, and thus training the obfuscation network through minimizing an accuracy loss of the surrogate network, and maximizing the first and second discriminant scores, wherein the universal discriminator has been trained by classifying de-identified images into positive or negative samples according to degrees of de-identification, generating discriminant scores determining degrees of de-identification for the samples through the universal discriminator, and training the universal discriminator by minimizing discriminator losses of the discriminant scores.