Protected-Region Image Processing for Privacy-Preserving Model Training

Solution Overview

Problem

Image analysis systems face challenges in accurately training machine learning models when regions containing sensitive information, such as faces or confidential data, undergo lossy processing, leading to significant feature value changes that render the images unsuitable for training.

Innovation Solution

An image processing device generates a second image through lossy processing on protected regions while preserving feature values by using image description information, allowing the creation of training images that maintain accuracy without compromising privacy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If lossy processing is applied to protection regions, then privacy protection is improved, but feature value accuracy deteriorates

Engineering Contradiction:

Improveprivacy protectionVSAvoidfeature value accuracy

Core Design Contradiction:

Object-affected harmful factorsVSMeasurement precision

Solution Approach 1:

The image is divided into protection regions and non-protection regions. Lossy processing is selectively applied only to the protection regions to protect privacy, while the non-protection regions maintain their original quality for training. This segmentation allows simultaneous privacy protection and feature value preservation where needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Image description information serves as an intermediary between the original image and the processed image. The system generates image description information from the original image, then uses this description to guide the lossy processing in protection regions while preserving essential features. This intermediary enables privacy protection without complete loss of feature information.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Object-affected harmful factors

If lossy processing is applied to protection regions, then privacy protection is improved, but training suitability deteriorates

Engineering Contradiction:

Improveprivacy protectionVSAvoidtraining suitability

Core Design Contradiction:

Object-affected harmful factorsVSReliability

Solution Approach 1:

The image is divided into protection regions and non-protection regions. Lossy processing is selectively applied only to the protection regions to protect privacy, while the non-protection regions maintain their original quality for training. This segmentation allows simultaneous privacy protection and feature value preservation where needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system changes the processing parameters dynamically based on region type. For protection regions, strong lossy processing parameters are applied to ensure privacy protection. For non-protection regions, original or minimal processing parameters are used to preserve training quality. This parameter adaptation ensures both privacy protection and training suitability.

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If feature values are preserved in protection regions, then training accuracy is improved, but privacy protection deteriorates

Engineering Contradiction:

Improvefeature value accuracyVSAvoidprivacy protection

Core Design Contradiction:

Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The image is divided into protection regions and non-protection regions. Lossy processing is selectively applied only to the protection regions to protect privacy, while the non-protection regions maintain their original quality for training. This segmentation allows simultaneous privacy protection and feature value preservation where needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Image description information serves as an intermediary between the original image and the processed image. The system generates image description information from the original image, then uses this description to guide the lossy processing in protection regions while preserving essential features. This intermediary enables privacy protection without complete loss of feature information.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20250322682A1Image processing device, image processing system, image processing method, and storage medium

Publication Date: 2025.10.16 CANON KK

AI summary

An image processing device sets one or more regions in a first image as a protection region, generates image description information as information expressing at least a portion of the first image, generates a second image based on lossy processing for the protection region in the first image, and performs control such that a training image used for training a learning model for image analysis processing is generated based on the second image and the image description information.