Tongue Body Segmentation And Texture Classification In Computerized Tongue Diagnosis

Tongue diagnosis is one of most important diagnosis methods in traditional Chinese medicine(TCM) and it has a long history in China. The TCM practitioner can analyze the physiological and pathological conditions of the patient by simply inspecting the color,shape, and texture of the tongue, making tongue diagnosis very promising for convenient and non-invasive diagnosis. However, traditional tongue diagnosis is a subjective skill which requires years of experience and practice. Moreover, for different practitioners,the diagnosis results may be inconsistent. The empirical and subjective factors affect the development of tongue diagnosis.With the development of image processing, several computerized tongue diagnosis systems have been reported. The systems are used to objectively acquire and analyze the information on the tongue body and assist physicians to make a diagnosis. The computerized tongue diagnosis system output the diagnosis results based on the tongue local features, which include tongue body contour, tongue coating, cracks and reflection points and so on. Tongue body contour detection affects tongue body segmentation and tongue shape analysis. Reflection points detection and removal can affect tongue humidity analysis and tongue color analysis. The tongue coating, papillary hyperplasia and cracks are the basis for the tongue diagnosis. And thus, the detection and expression on tongue local features are important for the computerized tongue diagnosis.This thesis make research on the tongue body segmentation, reflection detection and removal and texture expression and classification. To address existing problems in the detection and expression on tongue local features, this thesis proposes the following methods to solve the problems:1. Fast marching over the 2D Gabor magnitude domain for tongue body segmentation. Based on the characteristics of tongue body contour and Gabor filters, this paper uses Gabor filters to enhance the contour of tongue body. To solve the discontinuity on the tongue body, this paper propose fast marching method to connect the fragments of contour by initialization of stable segments. Finally, GVF snake model is used to obtain the tongue body segmentation. the experiment shows that the method is superior to the comparative tongue body segmentation methods.2. Stable segment guided interactive segmentation. Firstly, an automated tongue body segmentation method is used to obtain the initial tongue body. Secondly, edgedriven interactive method is introduced. Thirdly, this paper proposes three updating models to update tongue image, tongue edge image and stable segments. Finally, fast marching method and GVF snake model are used to obtain the final tongue contour. 3.Superpixel-based reflection detection method and weighted nuclear norm minimization(WNNM)-based reflection inpainting method. Since of the spit on the tongue, it is easy that highlight areas are formed on the tongue body. For the detection of highlight areas, this paper proposes a detection method based on superpixel, which can effectively detect sub-reflection areas and the highlight areas. For the inpainting method, this paper proposes a nonlocal method, image inpainting method based on WNNM. It proves that a stationary point can be obtained by solve WNNM problem, thus one result can be obtained. The experiment shows that this reflection detection method and the inpainting based on WNNM are superior to other methods.4. Tongue body texture classification based on tongue texture block: since of the background of tongue image, the reflection points and the randomness of location, scales and directions of tongue texture, tongue is complex. Therefor,it need a systemic method for the tongue body texture classification. and thus in this paper, we propose tongue body texture classification method based tongue texture block. In this method, first is preprocessing which includes tongue body segmentation, reflection discriminant, reflection detection and removal and tongue texture block detection. second is tongue texture block expression and classification. Finally, the results of tongue texture block classification is used to classify the tongue body texture. The experimental results show that the classification method based on tongue texture block is effective method for the tongue body texture classification.