Sweat Gland Extraction from Optical Coherence Tomography Using Convolutional Neural Network

Yilong Zhang,Xiaojing Li,Haixia Wang,Ruxin Wang,Peng Chen,Ronghua Liang
DOI: https://doi.org/10.1109/tim.2022.3223077
IF: 5.6
2022-01-01
IEEE Transactions on Instrumentation and Measurement
Abstract:As the Level 3 features of fingerprint, sweat pores have attracted attention in the field of fingerprint recognition and have been successfully applied to automatic fingerprint recognition systems. Traditional surface sweat pores become unclear or disappeared when the finger is contaminated, dried, or damaged. These unstable factors create major challenges in collecting sweat pores. Subcutaneous sweat glands belong to the internal tissues of fingers, which are stable and immune to external disturbances. This study investigated the extraction of subcutaneous sweat glands from fingertip volume data collected by optical coherence tomography (OCT). First, an improved multitask V-Net is proposed to extract subcutaneous sweat glands from OCT volume data. The network has an encoding path for features extraction and two decoding paths for extracting sweat gland boundaries and regions, respectively. The multitask scheme is designed to enhance the boundary and shape information of sweat glands and to prevent false extraction caused by interference from other tissues. Second, three mapping methods are proposed to address the problem of different spatial orientations of sweat glands. These three mapping methods, namely, global direct mapping (GDM), local direct mapping (LDM), and cylindrical fitting mapping (CFM), are used to map sweat glands to the surface fingerprint. Experiments are conducted in terms of sweat gland extraction, mapping, and matching. The qualitative and quantitative results show that the proposed network for sweat glands extraction outperforms other methods and that the LDM and CFM methods derive more accurate positions of sweat glands on the surface fingerprint than GDM. In the matching experiment, the equal error rate (EER) of dual-decoding V-Net (DDVN) reached 0.58%, which verified the recognition ability of sweat glands and the effectiveness of the proposed network.
What problem does this paper attempt to address?