Face Reconstruction From Anonymized Media Using Embedded Identity Features

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing facial recognition systems pose privacy concerns by accurately identifying individuals in media, and simply anonymizing faces does not allow for the reconstruction of original identities, hindering compliance with privacy regulations while preserving identity recognition.

Innovation Solution

A system using neural networks for adversarial training to embed and reconstruct facial features in anonymized media, minimizing perceptual differences and enabling recognition of original faces.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If faces are anonymized in media objects, then privacy protection is improved, but the ability to recognize and identify individuals is lost

Engineering Contradiction:

Improveprivacy threatsVSAvoididentity information

Core Design Contradiction:

Object-affected harmful factorsVSLoss of information

Solution Approach 1:

The patent extracts facial identity information from the original media object and separates it from the visual appearance. The facial features are extracted and embedded into the anonymized media object in an imperceptible manner, allowing identity recognition while maintaining privacy protection through the anonymized visual representation.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent embeds the extracted facial features within the anonymized media object structure. The identity information is nested inside the anonymized media container, making it imperceptible to human observers while still accessible to computational systems for recognition purposes.

Inventive Principle:

Principle #7Nested doll (Nesting)

2Reliability

If facial features are embedded in anonymized media objects, then identity recognition capability is improved, but perceptual differences become noticeable

Engineering Contradiction:

Improveface recognition accuracyVSAvoidperceptual quality

Core Design Contradiction:

ReliabilityVSEase of manufacture

Solution Approach 1:

The patent modifies only specific local regions of the anonymized media object where facial features are embedded, rather than altering the entire image. This localized modification approach maintains the overall perceptual quality of the anonymized media while embedding sufficient information for accurate face recognition.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The patent employs parameter optimization in the neural network training process to balance two competing objectives: maximizing face recognition accuracy and minimizing perceptual differences. By adjusting training parameters and loss function weights, the system achieves both reliable identity recognition and high perceptual quality.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12505251B2System and method for reconstruction of faces from anonymized media using neural network based steganography

Publication Date: 2025.12.23 DE IDENTIFICATION LTD

AI summary

A system and method for concealing and revealing a human face in media objects may include obtaining a first media object capturing an image of the face; employing a first unit to: extract, from the first media object, a set of features representing the face, and generate a second media object, by embedding the extracted features in an anonymized media object; and employing a second unit to recognize the face based on the second media object.