Research And Application Of Tongue Image Segmentation Oriented To TCM Diagnosis

With the development of medical science technology, basic diagnosis methods of Traditional Chinese Medicine (TCM) face demand of modernization. "Viewing" diagnosis, especially the tongue diagnosis is the important content of basic diagnosis methods in TCM. It can get the physiological function and pathology of the organism changes by means of observing the tongue and have very important meaning to make a definite diagnosis in earlier stage of a lot of diseases. However, traditional tongue diagnosis only depends on doctor's observation. It is difficult to make exact diagnosis because of the factor of personality. Therefore, combining the modern information technology and the clinic experience of TCM expert to realize the impersonality and standardization of tongue diagnosis of TCM is the only way of TCM development.In the tongue diagnosis, the tongue image of patient can be obtained by using digital instrument, for example, digital camera. Automatic interested region segmentation must be carries out before the tongue image processing. It makes the segmentation of tongue image an important part between image acquiring and image processing.Because the tongue image segmentation belongs to the color image segmentation research scope, this article first has carried on a summary to the existing color image segmentation algorithm. Then, after analyzing many existed algorithms of image segmentation and combining the characteristics of tongue image, the graph theory based image segmentation technique is selected to partition the tongue image.Based on the graph theory image division method was recently an international research hot spot. In this method, the image is mapped into a weighted undirected graph, and the pixels are considered as nodes, the segmentation result is obtained by minimum cut criteria. A survey on this technique is given in this paper. The fundamental conceptions on this technique are also introduced.Then, a new cut criteria has been proposed and formed a new image segmentation through converting the problem of minimizing of cut criteria to the solution of calculating the maximal flow of flow network. It has been used to segment the tongue image in TCM and found the result is very encouraging. In addition, the improvement of Normalized Cut which is a classical algorithm of graph theory based image segmentation is proposed, and adopt the local feature which is constructed by Local Binary Pattern. Its effectiveness is proved by experimental results. Finally, in order to improve the speed of segmentation, the algorithm of image segmentation by graph partition on histogram clustering is introduced, and the Region of interest is introduced also which is used in pretreatment of tongue image. The experimental results prove that the goal of segment image quickly is obtained.