Palmprint Features Fusion Recognition Based on Conformer and Gabor

In order to better extract the discriminative features of palmprint images and improve the accuracy of palmprint recognition, this paper proposes a palmprint feature fusion recognition algorithm based on Gabor and Conformer. First, Gabor filtering is applied to the palmprint region image to obtain the global distribution feature vector of image texture features. Meanwhile, a deep network feature extractor based on the Conformer neural network is constructed to obtain the local distribution feature vector of image deep features. Then, the kernel canonical correlation analysis (KCCA) method is used to extract the correlation features between the two groups of feature vectors as effective discriminative information, eliminate the information redundancy between the features, and obtain more accurate palmprint fusion features. Finally, a knowledge graph is used to pre‐classify the palmprint fusion features to reduce the feature search space, and the cosine distance classifier is used to recognise the palmprint. Simulation results show that the accuracy of the proposed method improves by about 15.02% compared with the basic features and about 3.92% compared with other algorithms, which proves the effectiveness of the proposed algorithm. The main contributions of this work lie in the joint encoding of global Gabor texture features and Conformer‐based local–global representations, the correlation‐aware fusion via KCCA, and the acceleration of large‐scale recognition through a knowledge‐graph‐based two‐stage matching mechanism.