Constitution Identification of Tongue Image Based on CNN

Abstract

The physique of traditional Chinese medicine (TCM) is the quality of our body and the tongue image is a manifestation based on the metabolism of the body. The constitution can be easily and objectively identified by the image of the tongue. In this paper, the classical convolution neural network (CNN) and gray level co-occurrence matrix, minimum enclosing rectangle and edge curve are used to extract the features of human tongue. Then different classifier are used to classify different constitution, and finally by comparing the accuracy and complexity of the two methods, we proposed a method constitution identification of TCM which is based on tongue images. The data set used in the experiment is provided and acquired by the Department of TCM in the hospital of Beijing University of Technology. In this paper, the accuracy of the three types of physique classification of Qi deficiency, damp heat and phlegm dampness is 63%, and the accuracy of traditional machine learning algorithm is respectively 30%,56% and 66%. It is of great significance to clinical, teaching and scientific research of TCM by making the most of the deep learning network and the auxiliary identification of physique.