Fault-Tolerant Multibiometric Recognition System Based on Neural Network

The multibiometric system ingeniously integrates multiple homogenous or heterogeneous biometric features, enhancing the precision and security of identity verification compared with single-biometric systems. However, this advantage comes with the challenge of increased costs, primarily due to the need for additional acquisition devices. Besides, biometric characteristics are vulnerable to damage in practical applications, thereby affecting recognition performance. To solve the above problems, this article innovatively proposes a fault-tolerant multibiometric system based on homogeneous five-finger vein images, effectively avoiding the extra costs associated with integrating heterogeneous biometric features. The system sets threshold conditions for successful recognition: when the number of correct finger vein images meeting the requirements reaches a preset standard. In other words, this system possesses fault tolerance, allowing for a certain number of users’ finger characteristics to be damaged. This establishes a flexible threshold recognition mechanism that enhances the system’s practicality and reliability. Specifically, an attention weight model (AWModel) was initially developed to dynamically adjust the weight strategy and effectively reinforce the significance of each sample in the overall analysis. Subsequently, a convolutional attention model (CAModel) was further designed to explore and extract in-depth features of the data. Finally, the features extracted by AWModel and CAModel were seamlessly integrated to form a more comprehensive and representative feature for classification. The experiment was conducted on three public databases, fully validating the system’s recognition performance and fault tolerance in multibiometric fields.