Image Feature Matching Through LDP Descriptor Perturbation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for image feature matching lack formal privacy guarantees, allowing raw descriptors to be used for recovering source images, posing significant privacy risks.
Innovation Solution
Implement a method that perturbs raw image descriptors using a subset selection mechanism and local differential privacy (LDP) protocols to generate perturbed descriptors, ensuring formal privacy guarantees, and matches these against a predefined descriptor database.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If raw image descriptors are shared with a centralized server for image matching, then image matching performance is improved, but privacy leakage risk increases as source images can be recovered from the descriptors
Solution Approach 1:
The patent applies preliminary action by performing descriptor perturbation before sharing with the server. The system replaces raw descriptors with perturbed versions that have been modified through subset selection and LDP protocols, preventing source image recovery while preserving matching utility for downstream tasks
Solution Approach 2:
The patent introduces an intermediary mechanism through the use of a predefined descriptor database and perturbation protocols. Instead of sharing raw descriptors directly, the system uses perturbed descriptors as an intermediate representation that maintains matching capability while blocking privacy attacks
2Reliability
If descriptor perturbation is applied to protect privacy, then privacy guarantees are improved, but image matching accuracy may deteriorate
Solution Approach 1:
The patent applies parameter changes by modifying the descriptor representation through controlled perturbation. The subset selection mechanism and LDP protocols transform descriptors in a way that preserves essential matching characteristics while removing identifiable information, achieving a balance between privacy and utility
Solution Approach 2:
The patent uses partial action by applying perturbation only to specific components of the descriptor while maintaining other critical features. The nearest neighbor replacement in the predefined database preserves matching capability by keeping descriptors within a controlled vocabulary that retains discriminative power
3Reliability
If a subset selection mechanism with large output domain is used, then privacy protection is improved, but computational complexity increases
Solution Approach 1:
The patent reduces computational complexity through preliminary action by pre-defining a descriptor database and replacing descriptors with nearest neighbors before the perturbation process. This preprocessing step limits the output domain size, making subsequent LDP operations more efficient while maintaining privacy guarantees
Data Source
AI summary
Systems and methods are provided for privacy-preserving image feature matching in computer vision applications, including receiving a raw image descriptor, and perturbing the raw image descriptor using a subset selection mechanism to generate a perturbed descriptor set that includes the raw image descriptor and additional descriptors. Each descriptor in the perturbed descriptor set is replaced with its nearest neighbor in a predefined descriptor database to reduce the output domain size of the subset selection mechanism. Local differential privacy (LDP) protocols are employed to further perturb the descriptor set, ensuring formal privacy guarantees, and the perturbed descriptor set is matched against a second set of descriptors for image feature matching.