Tongue Coating Classification Based on Multiple-Instance Learning and Deep Features

Tongue coating classification has long been a challenging task in Traditional Chinese Medicine (TCM) due to the fact that tongue coatings are multiform. Most existing methods make use of fixed location and handcrafted features, which may lead to inconstant performance when the size or location of the coating region varies. To solve this problem, our paper proposes a new tongue coating classification method. This method is mainly improved from two aspects: feature extraction and classification method. Complex tongue coating features extracted by Convolutional Neural Network (CNN) is used instead of handcrafted features, and a multiple-instance Support Vector Machine (MI-SVM) is applied to solve the uncertain location problem. Experimental results prove that our method shows significant improvements over state-of-the-art tougue coating classification methods.