Layer Segmentation of OCT Fingerprints With an Adaptive Gaussian Prior Guided Transformer

Optical coherence tomography (OCT) is a high-resolution, noninvasive imaging technology that can capture depth information 1-3 mm under the skin and obtain 3-D fingerprint structures. The accurate segmentation of 3-D fingerprint layers is conducive to improving the recognition accuracy and anti-spoofing ability of the fingerprint identification system. This article proposes a Gaussian prior guided transformer for layer segmentation of OCT fingerprints. First, the Gaussian transformer exploits the continuous layer structure information to achieve global modeling. Then, the features with different scales are fused. Finally, a spatial difference convolution module is introduced as complementary information to solve the problem of imprecise segmentation due to low contrast. The superiority is proven through experiments from the viewpoints of layer segmentation accuracy, fingerprint reconstruction quality, and fingerprint recognition performance. The MSE of the internal layer segmentation is 5.80, the average NFIQ score of the reconstructed internal fingerprint is 49.50, and the EER of the internal fingerprint recognition results reaches 1.60%.