Automatic Tongue Image Segmentation for Traditional Chinese Medicine Using Deep Neural Network

Automatic tongue image segmentation is a key technology for the research on tongue characterization in Traditional Chinese Medicine. Due to the complexity of automatic tongue image segmentation, the automation degree and segmentation precision of the existing methods for tongue image segmentation are not satisfied. To address the above problem, a method of automatic tongue image segmentation using deep neural network is proposed in this paper. In our method, an image quality evaluation method based on brightness statistics is proposed to judge whether the input image is to be segmented, and the SegNet is employed to train on the TongueDataset1 and TongueDataset2 to obtain the deep model for automatic tongue image segmentation. TongueDataset1 and TongueDataset2 are specially constructed for tongue image segmentation. The experimental results on TongueDataset1 and TongueDataset2 show that the mean intersection over union score can reach to 95.89% and 90.72%, respectively. Compared with the traditional methods of tongue image segmentation, our method can avoid the complicated process of extracting features manually, and has obvious superiority in the segmentation performance.