A method of tongue image segmentation based on kernel FCM

Abstract

Glossoscopy is an important part of Traditional Chinese Medicine (TCM). To analyze the tongue properties objectively, we need extract the tongue region from images. This paper presents a method to segment the tongue images based on kernel FCM (Fuzzy Cluster means). Firstly we pre-processed the tongue images by gray-level integral projection. Secondly the features were extracted to form a feature vector which contained texture, color, location and other information. Then, according to the feature vectors, pixels were clustering by kernel FCM whose parameters were decided by the proposed method. Finally, according to the pixels' neighbor connection theory, the tongue region was extracted. The results show that this method to segment tongue images is effective as its average accuracy reached upto 96.42%.