Selective Image Redaction via Synthetic Attribute Retention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image redaction technologies completely obscure faces and license plates, making them undetectable for analysis, which is not suitable for applications requiring recognition of non-identifiable features such as demographic analysis or vehicle type classification.
Innovation Solution
A system that uses machine learning models to selectively redact images by generating synthetic portions that retain essential attributes, allowing for detection without identification, enabling downstream analysis while maintaining privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If complete obfuscation (pixilation or blurring) is applied to faces and license plates, then privacy protection is improved, but detection and analysis capabilities are lost
Solution Approach 1:
The patent applies different levels of obfuscation to different regions within the same image. Critical identifying features (eyes, mouth, license plate numbers) receive stronger obfuscation while non-identifying structural features (face shape, vehicle type) maintain sufficient quality for detection. This regional differentiation resolves the contradiction by protecting privacy where needed while preserving detection capability where useful.
Solution Approach 2:
Instead of applying uniform complete obfuscation across entire faces or license plates, the patent applies partial obfuscation only to specific critical regions. This partial action approach maintains enough visual information for detection purposes while removing sufficient identifying details for privacy protection, thus resolving the information loss problem.
2Ease of manufacture
If uniform obfuscation is applied to all detected objects, then implementation simplicity is maintained, but analysis versatility is reduced
Solution Approach 1:
The patent implements dynamic obfuscation strategies that adapt to the specific analysis requirements. The system can adjust obfuscation intensity and region selection based on the type of analysis being performed (e.g., demographic analysis requires different obfuscation than vehicle classification). This dynamic adaptation provides versatility while maintaining a unified implementation framework.
Solution Approach 2:
The patent creates a universal obfuscation system that can serve multiple analysis purposes simultaneously. By preserving non-identifying structural features while removing identifying details, the same redacted image can be used for various analyses (demographic, behavioral, vehicle type classification) without requiring different processing pipelines, thus achieving both simplicity and versatility.
Data Source
AI summary
Selectively redacting an image by determining a set of attributes used by a machine learning model for an analysis, receiving image data detecting, by the one or more computer processors, a portion of the image data relevant to the analysis, the portion comprising at least some of the set of attributes, generating a synthetic portion from the portion, wherein the synthetic portion retains at least some of the attributes of the detected portion, replacing the portion with the synthetic portion, yielding redacted image data, and providing the redacted image data for analysis.