STUA-Net: A Fingerprint Reconstruction with Swin Transformer and Soft Collective Attention

Fingerprints play a vital role in person authentication and verification. To achieve accurate recognition, fingerprint images should contain 25 to 80 minutiae points, which define the unique characteristics of a fingerprint. However, due to various factors such as changes in the environment, the fingerprint structure can become corrupted, resulting in low-quality fingerprints. This corruption leads to a limited number of extractable minutiae points, making it challenging to establish the uniqueness of an individual. In this paper, we propose STUA-Net, a novel approach that incorporates Swin Transformer into the encoding and decoding layers to effectively map corrupted regions. Additionally, we introduce Soft Collective Attention to suppress the activation of relevant features. Our proposed method serves as a foundation for future research to improve recognition accuracy, particularly in scenarios involving low-quality fingerprints. It addresses an important problem in the field and contributes to the advancement of fingerprint recognition technology.