Obfuscation Network for Distinct Image Region Concealing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional image concealing technologies fail to effectively conceal personal information in big data, leading to exposure due to detection failures or excessive concealing that hinders information extraction for specific uses.

Innovation Solution

An obfuscation network is trained to perform distinct concealing processes on different regions of an image, generating an obfuscated image that is unidentifiable by naked eyes while maintaining recognizable features for learning networks, using a method that involves inputting training images into the network, applying varying degrees of concealing, and optimizing discriminator losses to ensure the obfuscated image is recognized as real by learning networks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional concealing technologies apply masks or blurs on detected regions of interest, then personal information exposure is prevented, but detection failures or missed personal information lead to exposure

Engineering Contradiction:

Improvepersonal information protection reliabilityVSAvoiddetection and concealing system complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The image is divided into multiple regions of interest (ROIs) based on different features (faces, texts, objects). Each ROI is independently identified and concealed with appropriate methods, allowing comprehensive coverage without requiring a single complex detection system to detect all types of personal information simultaneously.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different concealing methods and degrees are applied to different regions of interest based on their specific characteristics. For example, faces may use blur or distortion while texts may use pixelation, and the intensity of concealing varies according to the sensitivity and importance of each region, optimizing both protection effectiveness and information preservation.

Inventive Principle:

Principle #3Local quality

2Reliability

If excessive concealing process is performed on regions of interest, then personal information exposure is prevented, but extraction of information necessary for specific use becomes difficult due to information loss

Engineering Contradiction:

Improvepersonal information protection reliabilityVSAvoiduseful information loss

Core Design Contradiction:

ReliabilityVSLoss of information

Solution Approach 1:

The concealing intensity and method are locally optimized for each region of interest. Sensitive regions like faces receive stronger concealing while less sensitive regions receive milder treatment. This localized approach ensures adequate protection where needed while minimizing information loss in regions where full protection is not necessary.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

Instead of applying uniform excessive concealing to all detected regions, the system applies partial concealing only to the extent necessary for each specific region. The concealing strength is adjusted to be just sufficient for protection, avoiding over-concealing that would cause unnecessary information loss and hinder subsequent data utilization.

Inventive Principle:

Principle #16Partial or excessive action

3Ease of manufacture

If uniform concealing process is applied to entire image, then personal information protection is simplified, but important features necessary for training learning networks are lost

Engineering Contradiction:

Improveconcealing process simplicityVSAvoidtraining feature information loss

Core Design Contradiction:

Ease of manufactureVSLoss of information

Solution Approach 1:

The image processing is segmented into detection phase and concealing phase, with further segmentation of the image into multiple regions of interest. This structured segmentation provides a systematic framework that maintains relative simplicity while enabling differentiated treatment of different regions to preserve training features.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different concealing strategies are applied to different regions based on their importance for training. Regions containing critical training features receive milder concealing or selective concealing, while regions with sensitive personal information receive stronger protection. This local differentiation maintains usability for machine learning while ensuring privacy protection.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11023777B1Methods for training and testing obfuscation network capable of performing distinct concealing processes for distinct regions of original image and learning and testing devices using the same

Publication Date: 2021.06.01 DEEPING SOURCE INC

AI summary

A method for training an obfuscation network capable of performing distinct concealing processes for distinct regions of an original image is provided. The method includes steps of: a learning device (a) inputting a training image into the obfuscation network to generate an obfuscated training image by performing a 1-st to an n-th concealing process respectively on a 1-st to an n-th training region of the training image; (b) inputting the obfuscated training image into a 1-st to an n-th discriminator to respectively generate a 1-st to an n-th obfuscated image score on determining whether the obfuscated training image is real or fake, and inputting the obfuscated training image into an image recognition network to apply learning operation on the obfuscated training image to generate feature information for training; and (c) training the obfuscation network such that an accumulated loss is maximized, and an accuracy loss is minimized.