Federated Learning for Encrypted Image Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing encrypted image retrieval methods face challenges in supporting multi-source scenarios, achieving high accuracy, and reducing time consumption, particularly in large datasets and high-resolution images, while ensuring security and information theoretic security.
Innovation Solution
A multi-source encrypted image retrieval method based on federated learning and secret sharing, where a convolutional neural network is trained on double cloud platforms, and authorized users perform encrypted image retrieval using additive secret sharing, with protocols like Beaver's triplet-based secure multiplication and secure distance measurement to optimize feature extraction and comparison.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If homomorphic encryption is used to extract encrypted SIFT features, then security is improved, but time consumption increases to tens of hours
Solution Approach 1:
The patent divides the single cloud platform into two independent cloud platforms (C1 and C2), each holding shares of the encryption keys. This segmentation prevents either single platform from decrypting images alone, maintaining security while enabling efficient feature extraction through distributed computation that avoids the overhead of traditional homomorphic encryption operations.
Solution Approach 2:
The patent introduces an intermediary mechanism using secret sharing where the encryption key is split into shares distributed across two cloud platforms. This intermediary structure allows the system to perform encrypted computations without requiring a single platform to hold the complete decryption capability, thus reducing time consumption while preserving security through the use of additive homomorphic encryption on shared keys.
2Reliability
If garbled circuit technique is used in multi-party secure computing, then security is improved, but retrieval time increases to more than 10 seconds
Solution Approach 1:
The patent changes the cryptographic approach from garbled circuits to additive homomorphic encryption, fundamentally altering the parameter of encryption methodology. This parameter change enables more efficient computation by allowing arithmetic operations on encrypted data without decryption, reducing retrieval time from over 10 seconds to under 2 seconds while maintaining security through the mathematical properties of homomorphic encryption.
3Loss of time
If secret sharing technique is used to extract encrypted features, then retrieval time is reduced to less than 2s, but accuracy is still not high enough for large datasets
Solution Approach 1:
The patent transitions from traditional SIFT feature extraction to using pre-trained convolutional neural networks for feature extraction. This dimensional change in the feature space, leveraging deep learning representations, significantly improves retrieval accuracy on large datasets while maintaining the efficiency benefits of secret sharing and homomorphic encryption for the computation process.
4Device complexity
If single cloud platform is used for encrypted image retrieval, then device complexity is reduced, but information theoretic security cannot be guaranteed
Solution Approach 1:
The patent segments the cloud platform into two independent platforms (C1 and C2), each holding partial key information. This segmentation ensures that neither platform alone can decrypt images, providing information theoretic security through the secret sharing mechanism. While this increases system complexity compared to a single platform, the patent manages this through standardized protocols and interfaces for inter-platform communication.
Data Source
AI summary
Disclosed is a multi-source encrypted image retrieval method based on federated learning and secret sharing, including the following steps: S1. performing model training on a convolutional neural network of double cloud platforms based on federated learning, with an image owner joining the double cloud platforms as a coalition member; and S2. completing, by an authorized user, encrypted image retrieval based on additive secret sharing with the assistance of the double cloud platforms. The present disclosure provides a multi-source encrypted retrieval scheme based on federated learning and secret sharing, which simplifies the neural network model structure for retrieval by using federated learning, to obtain better network parameters. Better neural network parameters and a more simplified network model structure are achieved by compromising overheads on the image owner side, such that a better convolutional neural network can be used in encrypted image retrieval.