Anonymized Image Labeling via Obfuscation Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for labeling training data for neural networks face challenges in accuracy and resource efficiency, with mechanical labeling being less accurate and requiring significant resources, while manual labeling risks private information leakage, and existing anonymization methods are not entirely secure against data recovery.
Innovation Solution
A method involving transformation and obfuscation of original images to generate anonymized and obfuscated images, which are then labeled and combined to create a labeled image that is not identifiable by naked eyes but usable for neural networks, using techniques like shape transformation, noise addition, and convolutional layers to ensure privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to improve labeling accuracy, then labeling precision is improved, but private information leakage risk increases
Solution Approach 1:
The original image is divided into multiple anonymized images, each containing only specific labeling regions. Different labelers work on different anonymized images, preventing any single labeler from accessing the complete original image and thus reducing private information leakage risk while maintaining labeling accuracy.
Solution Approach 2:
An obfuscation network is introduced as an intermediary between the original image and the labelers. The network transforms the original image into anonymized images that preserve labeling regions while obscuring private information. The labelers interact with these anonymized images rather than the original image, solving the contradiction between accuracy and privacy protection.
2Object-affected harmful factors
If the original image is anonymized by blurring or adding noise to reduce private information leakage, then privacy protection is improved, but labeling regions cannot be properly recognized
Solution Approach 1:
Different regions of the image are treated differently: labeling regions are preserved with high quality for accurate recognition, while non-labeling regions are anonymized through blurring or noise addition. This local differentiation allows labelers to properly recognize labeling regions while private information in other regions remains protected.
Solution Approach 2:
The image is segmented into labeling regions and non-labeling regions. Only the labeling regions are retained in the anonymized images provided to labelers, while non-labeling regions are removed or anonymized. This segmentation ensures that labeling tasks can be performed accurately without exposing private information.
3Object-affected harmful factors
If anonymized data are distributed among multiple labelers to reduce private information leakage, then privacy protection is improved, but data can still be recovered if collected and combined
Solution Approach 1:
The obfuscation network serves as a secure intermediary that processes the original image to create anonymized images. This network applies cryptographic or transformation-based obfuscation that prevents reconstruction of the original image even if multiple anonymized images are collected. The intermediary ensures that the distribution and collection process itself cannot lead to data recovery.
Solution Approach 2:
The anonymization process applies parameter transformations (such as coordinate transformations, noise additions, or cryptographic encoding) to the image data that are irreversible or extremely difficult to reverse. These parameter changes ensure that even with multiple anonymized images, the original data cannot be reliably recovered.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method for producing a labeled image is provided. The method includes steps of: a labeling server (i) providing an image modifying interface to a user device to generate at least one anonymized image by anonymizing the original image except a specific labeling region among at least one labeling region, or generate at least one cropped image by cropping the labeling region, thus generating at least one transformed image by applying at least one transform function to the anonymized image or the cropped image, (ii) acquiring an obfuscated image by obfuscating the original image, (iii) acquiring at least one partial labeled image by allowing labelers to label the transformed image, and (iv) inversely applying the transform function received from the user device to the partial labeled image, thus generating at least one piece of adjusted partial labeling information and combining thereof with the obfuscated image to generate the labeled image.