CRCGAN: Toward robust feature extraction in finger vein recognition

Zhongxia Zhang, Zhengchun Zhou, Zhiyi Tian, Shui Yu

Abstract: Highlights•A novel CRCGAN is proposed for finger vein recognition. It integrates a classification function, overcoming CycleGAN’s limitations.•CRCGAN enables bidirectional mapping between finger vein images and features, extracting more compact features and addressing overfitting in traditional networks. Adversarial training enhances noise tolerance.•Experimental results show CRCGAN’s outstanding performance and robustness in finger vein recognition.