Research And Design Of Tongue Image Data Mining System Based On Mobile Terminal

Tongue diagnosis is an important part of TCM(traditional Chinese medicine) in the inspection, but its diagnosis depends on the doctor’s knowledge and working experience, so the diagnosis is subjective in a way. This will go against the promotion and development of the tongue diagnosis. With the development of mobile Internet and mobile terminal equipment, mobile phone has occupied first place in the usage of Internet. Therefore, to study a data mining system of tongue-diagnosis is more and more urgent, but also conforms to the requirements of the times.It is flexible and convenient for intelligent mobile terminal to acquire tongue images which include tongue color, tongue shape and so on, but it also prone to influence device diversity and environment interference etc.. In order to ensure the information is correct and reliable, this article did some research on the different color space models. At last we determined to carry out the research in the RGB and HSI model. We collected 88 sample images using three types of mobile phone, from which we extracted the value of R, G, B, H, S and I. After doing statistical analysis and the K- mean classification for the 264 sets of data, results show that the brightness of whiteboard is low, whiteboard seems like gray board because of the automatic exposure effect of mobile phone terminal. But the color feature vectors of R, G and B are generally consistent. Then we do image correction using this feature, experiment results show that the image color correction effect is good.Image segmentation is to provide an effective guarantee for the identification and correction of image color. The threshold segmentation method, K-means and Snake methods were studied. After doing some research on the effects of R, G, B, H, S and I for image segmentation, we found the value of red color in the H feature map is close to 0. So Otsu’s method was used for the white region segmentation, the experiment results show that the Otsu method can separate the white background. In order to eliminate the convex point of graphics, the image is processed by median filtering and smoothing filtering, and then the boundary of tongue is obtained by Sobel operator.. Then we used Snake to achieve tongue body segmentation, but the result still needs improve.Tongue image classification is the key of automatic recognition, this paper uses the SVM method to carry out research on tongue image classification. In the course of the study, this paper takes R, G, B, H, S and I as the basic characteristics of tongue. In order to obtain the best classification method, we fixed RBF, C and G parameter in the SVM method. Then the classifier was tested on the test samples, we found that C-SVC method is suitable for tongue image classification. This paper also explored how G parameter affects the classification result. We found, the correct rate is as high as 93.33% when G=0.01.After the research before, the overall system is designed and implemented. The design of the system used the C/S architecture. The main functions of the system are put forward: database module, data mining module, tongue diagnosis module, tongue medical etc.. This paper also realizes some functions of client software.