Double Branch Lightweight Finger Vein Recognition based on Diffusion Model

Aiming at the problems of high complexity, insufficient global information extraction and easy overfitting in finger vein recognition, a finger vein recognition method based on diffusion model is proposed. Firstly, finger vein images are generated according to the dataset by diffusion model, which is used to prevent overfitting; secondly, a streamlined convolutional neural network is used to form a two-branch lightweight backbone network with an improved multi-head self-attention mechanism, which can effectively reduce the complexity of the model; and finally, in order to maximally extract the image's overall information, the convolution is used to merge the extracted local and global features, and the recognition results are output. The algorithm can reach a maximum recognition rate of 99.78% on multiple datasets, while the number of references is only 2.15M, which further reduces the complexity of the algorithm while maintaining a high accuracy compared to other novel finger vein recognition algorithms as well as lightweight convolutional neural network models. As the first attempt in this field, it will provide new ideas for future research work.

@article{Tao2024,
  title     = {Double Branch Lightweight Finger Vein Recognition based on Diffusion Model},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {1},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Zhiyong Tao and Yajing Gao and Sen Lin},
  doi       = {10.14569/IJACSA.2024.0150147},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150147}
}