Biometric Recognition via Deep Hashing and Fuzzy Commitment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Biometric recognition systems face challenges in balancing template size, accuracy, and computational complexity, with existing methods suffering from privacy leakage and high computational complexity, especially in biometric cryptosystems.
Innovation Solution
The implementation of a Multi-spectral Palmprint Fuzzy Commitment Method based on Deep Hashing Code with Discriminative Bit Selection, which uses a deep hashing network to extract discriminative deep hashing codes from palmprint images across multiple spectrums, generating a biometric template and key through a fuzzy commitment scheme, thereby reducing storage cost and computational complexity while maintaining high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If biometric cryptosystem is used to strictly protect biometric features with one-way function, then privacy protection is improved, but computational complexity increases significantly
Solution Approach 1:
The patent transforms biometric features through parameter changes by applying orthogonal transforms (DCT, DWT, DFT) and non-linear transformations to convert original biometric data into transformed domains. This allows the system to protect privacy through irreversible transformations while maintaining computational efficiency, as the transformed features can be processed faster than traditional cryptographic methods.
Solution Approach 2:
The patent replaces complex cryptographic mechanical systems with signal processing-based protection mechanisms. Instead of using heavy cryptographic protocols, the system uses mathematical transforms and error-correcting codes to achieve both privacy protection and authentication, significantly reducing computational overhead while maintaining security.
2Quantity of substance
If biometric template size is reduced for efficient storage, then storage cost and computational complexity decrease, but recognition accuracy deteriorates
Solution Approach 1:
The patent extracts only the most discriminative and essential features from biometric data through feature selection and dimensionality reduction techniques. By taking out and retaining only the critical information needed for recognition, the system achieves compact template sizes without sacrificing recognition accuracy, as the extracted features capture the essential identity characteristics.
Solution Approach 2:
The patent applies parameter changes through quantization and encoding schemes that represent biometric features more efficiently. By changing the representation parameters (e.g., using compact codebooks, quantized vectors), the system reduces template size while preserving the discriminative information necessary for accurate recognition.
3Measurement precision
If original biometric features are used directly for recognition, then recognition accuracy is maintained, but privacy leakage occurs
Solution Approach 1:
The patent introduces transformed feature representations as intermediaries between original biometric data and the recognition system. These transformed features (through orthogonal transforms, non-linear mappings) serve as mediators that preserve recognition capability while preventing direct access to original biometric information, thus blocking privacy leakage pathways.
Solution Approach 2:
The patent creates transformed copies of biometric features that contain sufficient information for recognition but lack the direct identifying characteristics of original data. These copied and transformed representations enable the system to perform recognition operations without exposing actual biometric information, effectively decoupling recognition functionality from privacy risks.
Data Source
AI summary
The present disclosure provides a method for facilitating implementing biometric recognition. Further, the method may include receiving two or more biometric images of one or more biometric identifiers of one or more individuals from one or more devices. Further, the two or more biometric images may be in two or more spectrums. Further, the method may include analyzing the two or more biometric images using one or more deep hashing network models. Further, the method may include extracting two or more discriminative deep hashing codes from the two or more biometric images based on the analyzing. Further, the method may include generating a biometric template based on the two or more discriminative deep hashing codes. Further, the method may include generating a biometric key for the one or more biometric identifiers using a fuzzy commitment scheme based on the biometric template. Further, the method may include storing the biometric key.