Topic-Guided Image Region Evaluation for SNS Privacy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image processing systems for social networking sites (SNSs) fail to accurately protect privacy and security by manually designating regions for hiding personal information, leading to potential leaks of unintended information, and existing object detection methods do not align with user intentions for posting.

Innovation Solution

An information processing apparatus and method that estimates candidate regions and topics from image text information, evaluates regions based on their relevance to the topic, and generates images by hiding or emphasizing regions accordingly, using object detection algorithms and weakly-supervised learning to adapt to user preferences.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual processing is performed to designate regions for hiding personal information, then privacy protection can be achieved, but user labor and time are required

Engineering Contradiction:

Improveprivacy protectionVSAvoiduser labor

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system performs automatic region detection and evaluation without requiring user operation. The image processing apparatus autonomously identifies candidate regions, evaluates their relationship with the topic, and determines which regions to hide, thereby eliminating the need for manual user labor while maintaining reliable privacy protection

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The manual mechanical operation of designating regions is replaced by an automated information processing system using object detection algorithms and topic estimation. The system substitutes human judgment with computational evaluation of region-topic relationships, achieving both automation and reliable privacy protection

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If automatic object detection is performed based only on image information, then processing speed is improved, but accuracy in reflecting user intentions deteriorates

Engineering Contradiction:

Improveprocessing speedVSAvoidaccuracy of reflecting user intentions

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The system merges multiple information sources including image data, text information accompanying the image, and topic estimation results. By combining these different types of information, the system achieves both automated processing speed and accurate reflection of user intentions through comprehensive topic-based evaluation

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system uses topic estimation as feedback to guide the region evaluation process. The estimated topic provides contextual information that feeds back into the region selection algorithm, allowing the system to automatically adjust region selection based on the semantic meaning of the image content, thereby improving accuracy while maintaining speed

Inventive Principle:

Principle #23Feedback

3Reliability

If blindfolding is applied to faces only, then portrait rights are protected, but other personal information in the background may be leaked

Engineering Contradiction:

Improveportrait rights protectionVSAvoidprivacy leakage from background

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The system segments the image into multiple candidate regions and evaluates each region independently based on its relationship with the topic. This allows selective application of hiding processing to different regions (faces, background objects, etc.) according to their specific privacy risks, rather than applying a single uniform approach to the entire image

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system applies different privacy protection measures to different regions based on their local characteristics and topic relationships. Regions with high topic relevance are preserved while regions with low relevance (potential privacy risks) are hidden, achieving localized quality control that protects both portrait rights and background privacy

Inventive Principle:

Principle #3Local quality

4Shape

If trimming is performed based on object importance calculation, then image composition is improved, but user-desired segments may be eliminated or unnecessary segments shown

Engineering Contradiction:

Improveimage compositionVSAvoidalignment with user intentions

Core Design Contradiction:

ShapeVSMeasurement precision

Solution Approach 1:

The system uses topic estimation as feedback to guide trimming decisions. The topic provides semantic context that feeds back into the region evaluation, allowing the system to distinguish between important content and unnecessary background, ensuring that trimming reflects user intentions rather than just visual composition algorithms

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12443745B2Information processing apparatus and information processing method

Publication Date: 2025.10.14 SONY GROUP CORP

AI summary

Provided is an information processing apparatus that includes a region estimating unit estimate candidate regions of object detection from an image, a topic estimating unit that estimates a topic of the image on the basis of text information accompanying the image, a region evaluating unit that evaluates the candidate regions estimated by the region estimating unit on the basis of relationships with the topic estimated by the topic estimating unit, and an image generating unit that generates an image on the basis of evaluation results acquired by the region evaluating unit. The topic estimating unit described above estimates the topic on the basis of words to which tags are added.