Level set method based tongue segmentation in traditional chinese medicine

The correct segmentation of the tongue is the precondition to the quantification of tongue diagnosis in Traditional Chinese Medicine. The main method at present is by Snake model which uses the gradient of a tongue image as external energy to make the initial contour converge to the edge of the tongue body. Although this method has strong adaptive capability, it has a defect: for those images including the lower lip, the final contour often converges to the edge of the lip. To address the problem, a geometric model is proposed to correct the edges influenced by large concavity or sharp angle such as those with the lower lip. With this model, the initial contour does not include the whole lip. Then the initial contour is evolved toward the edge of the tongue body by Geodesic Active Contour Model after the Signed Distance Function is constructed. From the experiment results, the method is demonstrated to have good segmentation accuracy.