Image De-Identification Pipeline for Face and Text PII Obscuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques are inefficient in automatically detecting and removing personally identifiable information (PII) from images, including faces and text, which poses challenges in machine learning training and public-facing image usage.
Innovation Solution
A network-based system and method utilizing image and text recognition models to identify and obscure PII in images, employing blurring and masking techniques based on user-defined security settings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional techniques are used to detect and remove PII from images, then the process can be performed with simple tools, but the efficiency and effectiveness are insufficient
Solution Approach 1:
The system segments the PII detection and removal process into distinct functional modules: an image analysis system that detects faces and text, a text recognition system that identifies PII in text regions, and an image processing system that applies obscuration. This segmentation allows each module to specialize in specific tasks, improving overall efficiency while maintaining manageable system complexity through modular architecture.
Solution Approach 2:
The system introduces an intermediary text recognition component that bridges image analysis and PII detection. Instead of directly detecting PII from raw images, the system first identifies text regions, recognizes the text content, and then determines whether the recognized text constitutes PII. This intermediary step significantly improves detection effectiveness while the automated pipeline maintains high processing efficiency.
2Productivity
If manual methods are used to identify and remove PII, then accuracy can be high, but the processing time and productivity are low
Solution Approach 1:
The system implements self-service through automated machine learning models that perform image analysis, text recognition, and PII detection without human intervention. The image analysis system automatically detects faces and text regions, the text recognition system identifies PII content, and the image processing system applies appropriate obscuration. This automated pipeline achieves both high processing speed and maintained accuracy through trained algorithms.
Solution Approach 2:
The system incorporates feedback mechanisms where the text recognition system validates detected text regions and the image processing system adjusts obscuration based on detected PII types. The system uses feedback from each stage to refine subsequent processing, ensuring accurate PII detection while maintaining efficient automated processing throughout the pipeline.
3Measurement precision
If multiple models are used to recognize different types of PII, then the detection accuracy improves, but the system complexity increases
Solution Approach 1:
The system achieves universality through a multi-functional architecture where the image analysis system performs both face detection and text region identification, and the text recognition system handles multiple types of text-based PII (names, addresses, phone numbers, etc.). This multi-functionality allows the system to accurately recognize diverse PII types while reducing overall complexity by eliminating redundant components and streamlining the processing pipeline.
Data Source
AI summary
A computer system is provided and is programmed to: (1) receive a plurality of images; and/or (2) for each image of the plurality of images, the at least one processor is programmed to: (a) retrieve an image of the plurality of images; (b) execute at least one model to analyze the retrieved image to detect one or more individuals; (c) identify one or more items of text in the retrieved image; (d) analyze the one or more items of text to detect one or more personally identifiable items; (e) identify one or more items of text to obscure based upon one or more security settings; (f) update the retrieved image to obscure at least one of the one or more items of text to obscure and the one or more individuals; and/or (g) provide the updated image.