Automatic Tongue Image Segmentation Based on Thresholding and an Improved Level Set Model

Tongue diagnosis is an important and widely used diagnosis method in traditional Chinese medicine. Automatic tongue segmentation is crucial in the digital tongue diagnosis system. In this paper, an automatic tongue segmentation algorithm is proposed, which consists of two steps. In the first step, a thresholding method that combines color and gray level information is designed, which provides the initial contour needed in the second step. Next, an improved level set model based on geodesic active contour and Chan Vese model is put forward for boundary refinement. The weight function of the new model is adapted for a better balance of two sub-models. Experiment results show that the segmented area is more complete and closer to the real target with a lowest average ME value of 0.081 compared with other methods. The robustness of our algorithm is also verified by different tongue images in terms of shapes, lighting conditions and resolutions.