Machine Learning Based Tongue Image Recognition for Diabetes Diagnosis

Abstract

Tongue diagnosis of Traditional Chinese Medicine (TCM) is of great significance in the diagnosis of diabetes. To reduce the subjectivity of doctors in clinical diagnosis, this paper proposes a method for diabetic tongue image recognition based on machine learning. Specifically, the proposed method first transforms an image from RGB (red, green and blue) color space to HSV (hue, saturation, and value) color space, and the hue component is transformed into an image with higher difference degree. Secondly, image thresholding and morphological operation are performed on the image to obtain the initial tongue body region. Thirdly, the proposed method performs image thresholding and morphological operations on the value component to achieve image binarization, and retains the largest connected region as the final tongue body region. Fourthly, iterative image thresholding is used to further segment the tongue body region into tongue coating and tongue nature. Finally, the proposed method extracts tongue color features, and uses SVM (support vector machine) and KNN (k-Nearest Neighbor) to achieve image recognition based diabetes diagnosis. Experimental results show that SVM achieves higher diabetic image recognition accuracy than KNN.