Cross-Sensor Fingerprint Recognition Based on Style Transfer Network and Score Fusion

Chuan Cheng, Jiachen Yu, Linkai Niu, Zhicheng X. Cao, Heng Zhao

Abstract: With the emergence of various types of fingerprint sensors, the fingerprint images collected by different sensors are distinct from each other due to systemic deformation and the different imaging style. Most of the existing fingerprint recognition methods fail to consider the problem of cross-sensor fingerprint verification. This paper proposes a cross-sensor fingerprint recognition system based on style transfer and score fusion. The method uses a CycleGAN to unify the styles of fingerprint images from different sensors and combines ResNeSt-50 with a spatial transformation network (STN) to extract fixed-length texture features with two properties of domain alignment and spatial alignment. The texture features are used to calculate the Texture Comparison Score, which is fused with Minutiae Comparison Score to produce the final similarity score. Experiments are carried out on the MOLF database and the self-collected database by the Xidian University and show that the proposed method has achieved excellent results.

External IDs:dblp:conf/ccbr/ChengYNCZ23