Class Constraint-based Discriminative Features Learning Algorithm for Palm Print and Palm Vein Fusion Recognition

In recent years, hand-based multimodal biometrics has received a lot of attention in the field of pattern recognition due to its convenience, stability and inclusion of a wider range of recognition features. However, Some existing multimodal hand feature extraction methods extract the features of different modalities separately and then fuse them directly, mostly ignoring the correlation between the different modalities. In this paper, a multimodal biometric feature learning algorithm combining palm vein and palm print is proposed. The pixel difference vectors in multiple directions are first extracted by calculating the difference between each pixel in both modalities and its linear neighbour pixels. Then a simple and efficient feature learning algorithm is used, and the pixel difference vectors are projected into a low-dimensional binary code in a supervised manner. Finally, we perform a feature-level fusion of the learned binary features of palm veins and palm prints for multimodal hand feature recognition. The experimental results show that the method performs better than existing finger recognition methods.