Biometric Cryptosystem Using Neural Network Vector Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current biometric template protection schemes face challenges in achieving irreversibility, unlinkability, and renewability while maintaining recognition performance, particularly due to vulnerabilities in feature transformation and biometric cryptosystem schemes that are susceptible to security threats and cross-comparison attacks.
Innovation Solution
A biometric cryptosystem that employs random projection and artificial neural networks to map biometric vectors to secret vectors, ensuring irreversibility and unlinkability by projecting biometric feature vectors onto random subspaces and using error correction coding, thereby enhancing security and recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If feature transformation schemes (e.g., Biohashing, Bloom filter-based schemes) are used to protect biometric templates, then template security is improved, but the schemes are vulnerable to stolen-token scenario attacks and cross-comparison attacks
Solution Approach 1:
The patent introduces an intermediary transformation process using orthogonal matrices and non-linear functions between the original biometric template and the protected template. This intermediary layer (comprising orthogonal transformation followed by non-linear mapping) prevents direct access to the original biometric data, thereby defending against stolen-token attacks while maintaining template security
Solution Approach 2:
The patent transforms the biometric template by changing its parameters through orthogonal transformations and non-linear functions. The protected template is generated by applying these parameter transformations, which preserve the essential biometric information for recognition while altering the template structure to prevent cross-comparison attacks and enhance security
2Reliability
If other cancellable biometrics (CB) schemes are used to overcome stolen-token scenario vulnerabilities, then security against stolen-token attacks is improved, but the protected templates are exposed to cross-comparison attacks and biometric recognition performance is seriously degraded
Solution Approach 1:
The patent applies orthogonal transformations and non-linear functions to transform the biometric template parameters, creating a protected template that maintains recognition performance. The orthogonal transformation preserves the Euclidean distance structure, while the non-linear function adds complexity to prevent cross-comparison attacks without significantly degrading recognition accuracy
3Adaptability or versatility
If biometric cryptosystems with error correction coding are used to handle intra-user variation, then tolerance to template-probe differences is improved, but it is difficult to achieve unlinkability because any linear combination of error correction codewords may also be a codeword
Solution Approach 1:
The patent introduces an intermediary orthogonal transformation layer between the biometric template and the error correction coding process. This intermediary transformation ensures that even with linear combinations of codewords, the original biometric information cannot be recovered, thereby achieving unlinkability while maintaining tolerance to intra-user variations through the error correction capability
Data Source
Figure 1
Figure 1
Figure 2
AI summary
According to an aspect of the invention there is provided a method of enrolling a biometric sample in a cryptosystem scheme. The method comprises extracting biometric feature data from a biometric sample; generating a biometric vector from the biometric feature data; generating secret data; generating a secret vector from the secret data; combining the biometric vector and the secret vector using an artificial neural network, the artificial neural network configured to output a neural network model; and, storing the neural network model in a database, thereby enrolling the biometric sample in a cryptosystem scheme and facilitating subsequent verification through authenticating released secret data, released from the neural network model, whilst hiding the biometric feature data and the secret data. A method of verifying a biometric sample is also provided. Computer readable medium and systems are also provided.