Sweat Gland Enhancement Method for Fingertip OCT Images Based on Generative Adversarial Network

Optical Coherence Tomography (OCT) has been widely used for non-invasive cross-sectional imaging of biological systems, including the acquisition of fingerprint features such as sweat glands and pores on fingertips, which has gained attention in the field of fingerprint recognition. However, speckle noise in OCT images can cover sweat glands and blur tissue contours making extraction difficult. Although traditional denoising methods can remove speckle noise to some extent, they can cause unclear sweat glands and structural loss due to smearing and excessive smoothing. Therefore, this paper proposes a sweat gland enhancement method for fingertip OCT image based on generative adversarial network. A paired dataset generation strategy is proposed to extend a small number of manually enhanced groundtruth into a high-quality paired dataset. Then, an improved Pix2Pix image enhancement neural network is proposed, which adds a perceptual loss function to increase the weight of structural information in the original image. Experiments show that the proposed enhancement method can effectively remove speckle noise while restoring the spiral structure of sweat glands, improving the accuracy of sweat gland segmentation and extraction.