Scholarlink

In the world of biometrics, finger vein biometrics is one of the hot research areas and there is a lot of scope for innovative research. But the availability of finger vein data is very limited as only couple of publicly available datasets are present to experiment with and the number of samples are not that high for a deep learning framework to be experimented with. In this paper we have focused on utilizing generative AI to create finger vein images through hybrid approaches of using two different conditional generative adversarial network (cGAN) that can take care of the multi domain and multi class scenarios as well as handle the semantic features of an image while creating a n image along with labels. Further to enhance the performance of finger vein recognition we have experimented with utilizing transfer learning by leveraging the state-of-the-art predefined models and with the synthetic data generated by our hybrid approach of using the cGANs, we utilized the transfer learning to further improve the accuracy of the model to around 99% which is very good compared to some of the research done by various researchers in the similar field. The complete explanation of the proposed hybrid framework of using transfer learning and generative AI has been mentioned in this paper along with a comparative study.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid Fusion Model Using Advanced Conditional Adversarial Networks and Transfer Learning for Finger Vein Recognition

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Kiranjit Pattnaik,
  • Partha Pratim Sarangi,
  • Bhabani Shankar Prasad Mishra

摘要

In the world of biometrics, finger vein biometrics is one of the hot research areas and there is a lot of scope for innovative research. But the availability of finger vein data is very limited as only couple of publicly available datasets are present to experiment with and the number of samples are not that high for a deep learning framework to be experimented with. In this paper we have focused on utilizing generative AI to create finger vein images through hybrid approaches of using two different conditional generative adversarial network (cGAN) that can take care of the multi domain and multi class scenarios as well as handle the semantic features of an image while creating a n image along with labels. Further to enhance the performance of finger vein recognition we have experimented with utilizing transfer learning by leveraging the state-of-the-art predefined models and with the synthetic data generated by our hybrid approach of using the cGANs, we utilized the transfer learning to further improve the accuracy of the model to around 99% which is very good compared to some of the research done by various researchers in the similar field. The complete explanation of the proposed hybrid framework of using transfer learning and generative AI has been mentioned in this paper along with a comparative study.