Image Data Anonymization via Frequency Domain Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge is to securely protect the privacy of image data, particularly face image data, from unauthorized access and malicious theft in third-party processing environments, where data security is unpredictable, and original image data can be reconstructed from stolen feature data, posing a threat to services like secure payments.
Innovation Solution
An image data processing method that performs anonymization using frequency domain transforms and data augmentation techniques, such as mixup data augmentation, to create augmented anonymized image data, making it difficult to reconstruct the original data and ensuring privacy protection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If original image data is provided to third-party device for processing, then service processing can be performed, but data security is compromised due to unpredictable security level of third-party device
Solution Approach 1:
The patent segments the image data processing into two distinct parts: (1) anonymization processing performed locally on the user's device to protect privacy, and (2) feature extraction and service processing performed on the third-party device using the anonymized data. This segmentation allows service processing to occur externally while keeping sensitive original image data secure on the user's device.
Solution Approach 2:
The patent introduces anonymized image data as an intermediary between the original image data and the third-party device. This intermediary form contains sufficient information for service processing (feature extraction) while removing personally identifiable information, thus enabling service processing capability while protecting data security.
2Measurement precision
If feature extraction is performed on original image data, then recognition accuracy is improved, but data privacy is threatened as stolen feature data can be used to reconstruct original data
Solution Approach 1:
The patent performs anonymization processing as a preliminary action before feature extraction. By removing or obscuring personally identifiable information from the image data before it undergoes feature extraction, the system ensures that even if feature data is stolen, it cannot be used to reconstruct the original identifiable image data, thus protecting data privacy while maintaining recognition accuracy.
3Reliability
If image anonymization processing is performed before providing original image data to third-party device, then data privacy is protected, but processing complexity increases
Solution Approach 1:
The patent implements self-service by performing the anonymization processing on the user's own device rather than relying on the third-party device for security. This allows the user to control the anonymization process and ensures that original image data never leaves their device in identifiable form, protecting data privacy while keeping the third-party device's security requirements simpler.
Data Source
AI summary
Implementations of the present specification provide an image data processing method, an image data recognition method, a training method for an image recognition model, an image data processing apparatus, an image data recognition apparatus, and a training apparatus for an image recognition model. During image data processing, data anonymization processing is performed on image data based on frequency domain transform to obtain anonymized image data of the image data. The obtained anonymized image data includes a subgraph data set. Each subgraph data in the subgraph data set corresponds to a different frequency. Then, image blending processing is performed on the obtained anonymized image data based on data augmentation, to obtain augmented anonymized image data. In some implementations, graph size alignment processing is performed on each subgraph data in the augmented anonymized image data, so that a size of each subgraph data after the graph size alignment processing is same as a size of the image data in original form.