Obfuscated Image Generation for Neural Network Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing learning networks trained with obfuscated images may not operate correctly when inputted with original images due to inevitable correlations between obfuscated parts and labels, leading to incorrect responses.
Innovation Solution
A method involving a labeling device that detects privacy-related regions, adds dummy regions, and obfuscates these regions using algorithms or networks to generate training images, ensuring the learning network can operate correctly with both obfuscated and original images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If obfuscated images are used as training images, then privacy protection is improved, but the learning network cannot operate correctly with original images due to correlations between obfuscated parts and labels
Solution Approach 1:
The patent applies preliminary action by adding dummy regions to the original image before obfuscation. This preparatory step creates a modified training image where dummy regions are strategically placed in areas that do not contain important features. When these dummy regions are subsequently obfuscated, they do not create harmful correlations with the labels, allowing the network to learn effectively from obfuscated images while maintaining correctness on original images.
2Reliability
If conventional obfuscation techniques are applied to training images, then privacy-related regions are concealed, but correlations between obfuscated parts and labels are created
Solution Approach 1:
The patent applies local quality by differentiating between regions that should be obfuscated and regions that should remain intact. Important feature regions are preserved without obfuscation, while dummy regions are selectively obfuscated. This localized approach ensures that privacy protection is applied only where needed, without creating spurious correlations that would harm the learning network's ability to process original images correctly.
3Reliability
If all regions are obfuscated to ensure privacy, then privacy protection is improved, but the learning network loses ability to learn from important features
Solution Approach 1:
The patent applies segmentation by dividing the image into distinct regions: privacy-sensitive regions that require obfuscation, important feature regions that must be preserved, and dummy regions that can be safely obfuscated. This segmentation allows selective obfuscation of only the dummy regions while leaving important features intact, thereby maintaining both privacy protection and feature preservation for effective learning.
Data Source
Figure 1~2
Figure 3A~3B
Figure 3C~3E
AI summary
A training image to be used in training a learning network is generated. The method of generating the training image includes steps of: (a) a labeling device, in response to acquiring an original image, (i) inputting the original image into an image recognition network to detect privacy-related regions from the original image, (ii) adding dummy regions, different from the detected privacy-related regions, onto the original image, and (iii) setting the privacy-related regions and the dummy regions as obfuscation-expected regions which represent regions to be obfuscated in the original image; (b) the labeling device generating an obfuscated image by obfuscating the obfuscation-expected regions; and (c) the labeling device labeling the obfuscated image to be corresponding to a task of the learning network to be trained, to thereby generate the training image to be used in training the learning network.