Facial Image Normalization Using Synthetic Data for Privacy-Safe Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing facial recognition technologies face challenges in balancing the need for effective public safety with protecting individual privacy, particularly in the normalization of biometric data for identification documents, which often involves the use of personally identifiable information (PII) that can be misused.

Innovation Solution

A method and system for generating normalized facial images from non-normalized images using a neural network, removing background artifacts, and controlling output devices based on these normalized images, while minimizing the use of actual personal identifiable information (PII) by employing synthetic data and machine learning techniques.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional facial recognition systems use actual PII for normalization, then recognition accuracy is improved, but privacy protection deteriorates

Engineering Contradiction:
Improvefacial recognition accuracyVSAvoidprivacy risk
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent uses synthetic facial images generated by GANs as copies of real facial data for training and normalization. These synthetic images replicate the statistical properties and visual characteristics of real faces without containing actual PII, thereby maintaining recognition accuracy while eliminating privacy risks associated with using real facial data.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent introduces synthetic facial images as an intermediary between real facial data and the facial recognition system. This intermediary layer allows the system to learn from data that mimics real faces without directly processing sensitive PII, thus resolving the contradiction between accuracy and privacy protection.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If facial normalization processes are implemented, then recognition reliability is improved, but the risk of PII misuse increases

Engineering Contradiction:
Improverecognition reliabilityVSAvoidPII misuse risk
Core Design Contradiction:
ReliabilityVSObject-generated harmful factors

Solution Approach 1:

The patent replaces actual PII used in normalization processes with synthetic facial images that copy the essential features needed for reliable recognition. The GAN-generated images provide the necessary training data for normalization algorithms without exposing real individuals' facial information, thus maintaining reliability while preventing PII misuse.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent uses disposable synthetic data that can be freely generated and discarded without concern for privacy protection. These synthetic facial images serve as temporary training materials that fulfill the normalization requirement without the long-term privacy concerns associated with storing and processing real PII.

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

3Measurement precision

If more actual PII is collected for training, then system accuracy is improved, but privacy protection and security deteriorate

Engineering Contradiction:
Improvesystem accuracyVSAvoidprivacy protection
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent generates large volumes of synthetic training data through GANs, copying the diverse characteristics of real facial data without collecting actual PII. This approach provides sufficient training material for high system accuracy while maintaining privacy protection, as the synthetic data can be generated indefinitely without involving real individuals.

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The patent transforms the nature of training data from real PII to synthetic representations, fundamentally changing the parameter of data origin while preserving the statistical properties needed for accurate recognition. This parameter change allows unlimited data collection for training without the privacy risks inherent in collecting real facial information.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4411573B1Facial recognition systems and methods and associated automated controls
Publication Date: 2026.03.11 AMADEUS SAS
  • EP4411573B1 patent drawingFigure 1
  • EP4411573B1 patent drawingFigure 2
  • EP4411573B1 patent drawingFigure 3

AI summary

The present specification provides a system and method to generate passport-like images. The passport-like images can be generated from synthetic images, without use of personal identifiable information (PII). The passport-like images can be used for different applications such as to provide a machine learning dataset that can be used to control output devices based on a comparison of a real-time photograph of an individual with a passport photo of the individual.