Research On Computerized Classification Of Tongue Color In Traditional Chinese Medicine

Tongue Diagnosis is one of the most clinically valuable methods in Traditional Chinese Medicine. With the rapid development of computer science and technology these years, it is inevitable that traditional Tongue Diagnosis should be computerized. This dissertation thus aims at helping the development of computerized Tongue Diagnosis and researching on the methodology of color training and classification of tongue images.The main contributions of this dissertation include: designing the framework of pixel-based tongue color classification system; proposing the 2-stage FCM algorithm and solving the tongue color model construction problem in pixel-based tongue color classification system; proposing the dynamically local KNN algorithm for tongue substance and tongue coating color classification, and improving the system speed greatly; proposing the 12-dimension feature vector of color ratio and applying it to color classification of tongue image; doing research on the automatic diagnosis of diseases and symptoms using color, texture and shape information.Firstly, this dissertation studies on the detection algorithms of highlight and shadow of tongue images in the preprocess flow. An adaptive area/lightness-based highlight detection algorithm and an improved shadow detection algorithm eliminating marginal shadow are proposed. By comparing with other algorithms, we find that the experiment results of the proposed algorithms are better.Secondly, aiming at the necessity of pixel-based tongue color classification system, this dissertation proposes a 2-stage FCM algorithm. It applies the techniques of Pattern Classification and makes the better automation of tongue color model construction true. The experiment shows that the result is satisfied.Thirdly, combining with the characteristic of the tongue color data, this dissertation proposes a dynamically local KNN algorithm for tongue substance and tongue coating color classification, which is more suitable for training color dataset of large scale. The algorithm is proved speedy and efficient in the experiment.Finally, by utilizing the 12-dimension feature vector of color ratio and other...