Data Augmentation Using Image-to-image Translation for Tongue Coating Thickness Classification with Imbalanced Data

Mingxuan Liu, Yunrui Jiao, Hongyu Gu, Jingqiao Lu, Hong Chen

Abstract: Tongue diagnosis is widely used in traditional Chinese medicine diagnosis. The classification of tongue coating thickness is one of the most important tasks in tongue diagnosis. However, data imbalance imposes challenges when using deep learning methods for tongue coating thickness classification. In this paper, we propose a data augmentation method using image-to-image translation to solve the data imbalance problem. First, we use an image-to-image translation model based on generative adversarial networks (GANs) to translate thick and thin tongue coating images into each other, then we train the classification model using synthetic images together with real images. Finally, the trained classification model is used to classify the thickness of tongue coating. With our data augmentation method, the classification performance yields 0.92 accuracy and 0.922 F1-score, which is 3.37% and 3.95% higher than that with re-sampling method respectively.