Fissured Tongue Image Recognition Based on Support Vector Machine

Abstract

Tongue diagnosis is a primary method of traditional Chinese medicine (TCM) diagnosis, and the identification of fissured tongue is one of the important contents of tongue diagnosis, since fissured tongue always reflects some diseases. In this paper, the recognition of fissured tongue is studied. Firstly, the images of fissured tongue and non-fissured tongue were preprocessed by median filtering, histogram averaging and tongue segmentation. Because there are obvious texture and gray gradient differences between fissured and non-fissured areas in tongue images, local binary pattern (LBP), histogram of oriented gradient (HOG) and haar-like feature extraction were applied to tongue images respectively to get the input vectors. Then support vector machine (SVM) with four different kernel functions were respectively applied to train the classifiers and five-fold cross validation was adopted to get the average accuracy, precision and recall of the classification model. The results show that LBP features with linear kernel function can get the best classification effect, among which the accuracy rate is 97.72%, the precision rate is 97.46%, and the recall rate is 98.06%. This research lays a foundation for further fissure extraction in fissured tongue, and further facilitates the development of intelligence and automation of tongue diagnosis in the field of traditional Chinese medicine.