Image Redaction Using AI Detection of Sensitive Personal Traits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of protecting sensitive information, such as personal data, in image data shared through social networks or cloud storage, as existing technologies fail to effectively identify and remove non-object-based sensitive information like emotions or health conditions.
Innovation Solution
An electronic device employs a sensitive information detection model to identify areas in image data containing sensitive information, followed by an image reconstruction model to modify and remove these areas, ensuring secure sharing and storage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If image data is shared through social networks or cloud storage, then accessibility and usability are improved, but privacy and security deteriorate due to exposure of sensitive information
Solution Approach 1:
The patent applies preliminary action by detecting and removing sensitive information from images before they are shared through social networks or cloud storage. The sensitive information detection model scans images in advance to identify sensitive areas (such as personal data, health conditions, or private information), and the image reconstruction model removes these areas before the image is made accessible to others, thus preventing privacy exposure while maintaining sharing functionality
Solution Approach 2:
The patent introduces an intermediary mechanism consisting of two AI models: the sensitive information detection model and the image reconstruction model. These models act as intermediaries between the user's original image and the shared image, automatically identifying and removing sensitive information without requiring user intervention, thus resolving the conflict between ease of sharing and privacy protection
2Reliability
If existing technologies are used to protect sensitive information, then some basic filtering is provided, but non-object-based sensitive information like emotions or health conditions cannot be effectively identified or removed
Solution Approach 1:
The patent applies parameter changes by transforming the detection approach from object-based to feature-based analysis. The sensitive information detection model analyzes multiple parameters including facial features, skin tone variations, eye color, and other physiological indicators to detect non-object-based sensitive information such as emotions, health conditions, and genetic traits. This enables the system to identify sensitive information that traditional object-detection methods cannot capture
Solution Approach 2:
The patent replaces traditional mechanical or rule-based filtering systems with AI-based detection and reconstruction models. These models use machine learning algorithms to automatically detect and remove sensitive information, providing adaptability to various types of sensitive data including emotions and health conditions, rather than relying on predefined rules that can only handle object-based information
Data Source
AI summary
A method of removing sensitive information from image data by an electronic device is provided. The method includes obtaining first image data, determining whether the first image data includes at least one piece of sensitive information, when the first image data is determined as including the at least one piece of sensitive information, identifying a first area related to the at least one piece of sensitive information in the first image data, and generating second image data by modifying the first area.