Abstract:

We propose a mean shift based tongue segmentation algorithm for the characteristics of a tongue image. It initially smooths an input tongue image by mean shift method, which can effectively suppress the interference of tongue crackle and color block noise on tongue segmentation. It then implements a clustering process based on color domain and space domain proximity. A tongue image is segmented according to the clustering result. Experimental results show that segmentation results of different tongue images satisfy the diagnosis requirements of a traditional Chinese medicine (TCM) doctor. The algorithm can better implement tongue segmentation for a noise blended image.

Key words: tongue segmentation, mean shift, kernel method, clustering