Abstract:

[Objective] In order to improve the security and accuracy of biometric recognition technology, this paper proposes an algorithm that integrates three modalities of iris, face, and periocular at the score level. [Methods] Firstly, the algorithm uses a lightweight convolutional neural network as the feature extractor, which calculates the cosine similarity between feature vectors as the matching score between different objects. Secondly, the good point set initialization is used to enhance the population diversity of the Coati Optimization Algorithm. Levy flight is added in the exploration phase to improve the global search capability. The improved Coati Optimization Algorithm is used to solve the best parameters of the three modal scores under predefined fusion rules. Finally, the Schweizer operator is used to perform fuzzy inference on different parameter combinations, and the minimum membership degree method is used to defuzzify and obtain the optimal score fusion rules and their parameters. [Results] A homogenous facial multimodal dataset was constructed from the CASIA-IrisV4-Distance dataset for comparative experiments. The experimental results show that compared with the baseline model, the equal error rate (EER) decreases by 0.89%, the false mismatch rate (FNMR) decreases by 3.32% when the false matching rate (FMR) is 10^-5, and the discriminative index improved by 0.61. Compared to four optimization algorithms, this algorithm has higher recognition accuracy. [Conclusions] It can be seen that the algorithm proposed in this paper has achieved good recognition performance in multimodal score layer fusion.

Key words: multimodal fusion, coati optimization algorithm, biometric recognition, score layer fusion