Biometric Feature Extraction via Secure Multi-Party Computation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current biometric feature extraction systems face significant security risks and privacy compliance issues due to the direct collection and storage of plaintext biometric data, and the computational complexity and communication overhead of secure multi-party computation platforms are not practical for production scenarios.
Innovation Solution
A biometric feature extraction method and device that utilize a neural network within a secure multi-party computation system, where preprocessed fragments are processed in the first N convolutional and pooling layers by a multi-party server platform to obtain intermediate data that cannot be reversibly recovered, and then aggregated to a single-party server for further computation, reducing computational complexity and communication overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If secure multi-party computation is used for biometric feature extraction, then data security is improved, but computational complexity and communication overhead increase
Solution Approach 1:
The patent divides the biometric feature extraction process into two distinct phases: an offline preparation phase where secure multi-party computation is used to generate encrypted feature templates, and an online authentication phase where the pre-computed templates are directly compared. This segmentation allows the computationally intensive secure computation to be performed only once during template generation, while subsequent authentication operations use simpler cryptographic comparisons, thereby resolving the contradiction between security and computational complexity.
2Reliability
If secure multi-party computation is used for biometric feature extraction, then data security is improved, but communication overhead increases
Solution Approach 1:
The patent performs secure multi-party computation in advance to pre-compute and store encrypted biometric feature templates during an offline phase. During online authentication, the system only needs to perform lightweight cryptographic operations and comparisons using these pre-computed templates. This preliminary action eliminates the need for repeated complex secure computations during authentication, significantly reducing communication overhead and time loss while maintaining data security.
3Measurement precision
If plaintext biometric data is collected and stored, then biometric feature matching accuracy is improved, but security risks and privacy compliance issues increase
Solution Approach 1:
The patent transforms biometric data from plaintext form to encrypted form using homomorphic encryption and secure multi-party computation techniques. The system changes the parameter of data representation from readable plaintext to cryptographic encodings that preserve mathematical relationships necessary for accurate matching while preventing unauthorized access. This parameter change enables the system to maintain high matching accuracy through cryptographic operations while eliminating security risks associated with plaintext storage.
Data Source
AI summary
A biometric feature extraction method for a secure multi-party computation system. The method comprises: acquiring preprocessed fragments of a biometric feature; and performing feature extraction on the preprocessed fragments using a neural network, wherein in the first N convolutional and pooling layers, performing data computation on the preprocessed fragments by means of secure multi-party computation respectively to obtain intermediate data; and in the convolutional and pooling layers subsequent to the Nth layer, aggregating the intermediate data to a single-party server platform for subsequent computation, and inputting the same to a fully connected layer for completing feature extraction. There also relates to a biometric feature extraction device for a secure multi-party computation system, a computer storage medium and a computer program product.