Research On Features And Classification Evaluation Of Tongue Color In Traditional Chinese Medicine

Tongue diagnosis is one of important approaches for getting significant evidences indicating the patient's health conditions in traditional Chinese medical diagnosis; it plays an important role in medical care diagnosis for a long time. With the rapid development of computer science and technology these years, it is inevitable that traditional Tongue Diagnosis should be computerized.The purpose of our research is to combine modern sciences and technologies, such as image processing, pattern recognition, artificial intelligence and computing technology, etc., with the knowledge and experience of traditional Chinese medical diagnosis, and to raise the level of objectiveness, quantification and autoimmunization of traditional Chinese medical diagnosis. This paper aims at researching on features and classification evaluation of tongue color.Firstly, colorful apperceive rule of color imaging system and characteristic and transformation each other between color spaces is researched. Original tongue images in RGB are transformed to other color spaces. The stability of some familiar color space under different brightness is analyzed. These provide theory for selecting color space, tongue segmentation and color features.Secondly, an automatic method of tongue segmentation is presented base on intensity transformation. Tongue image which include intensity information is transformed to proper intensity transformation, histogram equalization. Then threshold is determined by Otsu and tongue body region is picked up by marking pixel. Finally, tongue image is presented by mathematical morphology. This method is counted simply and implemented easily. Thirdly, a method for separating coats and substance and picking up features based on FCM algorithm is developed. It provides a lot of tongue color data for color classification.Finally, the performance of five-familiar tongue color classification is evaluated, such as Bayes, BPNN, SVM, LDC, K-NN, base on different color spaces. Comparing 9-3D color-spaces and 18-2D chrominance planes, the system of tongue-color classification performance are built to choose proper color-space to detect character and choose better classify arithmetic. According to the system, the basis is provided for selecting proper color space and much better classification arithmetic.