Research On Tongue Image Analysis Of Chinese Medicine In Kidney Disease Recognition

Chronic kidney disease(CKD) has a high coverage in chronic disease.The treatment of chronic kidney disease is a long process, it can be divided into different stages of kidney disease based on the patient's current symptoms, Chinese medicine, the tongue can reflect the state of the patient's body organs lesions in Chinese medicine for treatment of tongue diagnosis is particularly important, in the judgment and treatment of chronic diseases can play a good supporting role, it is of great significance for tongue inspection of information.In this thesis,we studies the relationship between the tongue and kidney disease base on tongue image,finally realize CKD prediction based on tongue image. The main work includes tongue image segmentation, feature extraction and CKD prediction. The following work is illustrated in this thesis:1. A novel tongue image segmentation based on redress Grab Cut algorithm is proposed. In contrast to a single Grabcut image segmentation algorithm has lower time complexity,and the time complexity problem of Grab Cut algorithm is solved. A novel tongue image segmentation based on SLIC and similar region merging algorithm is proposed. A novel tongue image segmentation based on SLIC and Grab Cut algorithm is proposed,the problem of high resolution make a high time complexity in Grab Cut is solved.2. In order to quantify the image color information, the tongue image pixel is divided according to the similarity of the 12 color feature points, the feature set is calculated for each color in the image proportion, ultimately tongue image quantified as 12-dimensional properties; introduction of body tongue zoning approach taken in each area of the image 64 x 64 pixels in size, calculate the energy gray symbiotic moments, the tongue image quantized to 8-dimensional texture attributes. Each tongue final image is quantized to 20-dimensional properties.3. Classification based on heterogeneous integrated classifier ensemble algorithm was used to improve the Classification accuracy, respectively, SVM, KNN, Bayes, DTree four single classification algorithm classifier training, according to the prediction error rate of different classification weighting the different classifiers, and in accordance with each classifier weights weighted vote based on the single classification on improving the prediction accuracy rate of kidney disease; increased more than 5% of the classification algorithm based on prediction accuracy of weighted voting.