Tongue shape classification integrating image preprocessing and Convolution Neural Network

Abstract

Tongue diagnosis is one of the most important parts in “inspection diagnosis” of Traditional Chinese Medicine (TCM). Observing tongue shape can help to understand the changes in human body and thereby to estimate the illness. This paper presents a method of recognizing tongue shapes based on Convolution Neural Network. The proposed method enhances the features of tongue images with preprocessing to ensure the data suitable for tongue shape binary classification. In view of the special texture and outline of tongue, the whole tongue images of dot-sting tongue and fissured tongue is transformed by Gabor filter, and the tooth-marked are processed by boundary detection approach. CNN is adopted because it has achieved remarkable results in computer vision and pattern recognition, and the model training through neural network coincides with the Chinese medicine dialectics through experience. Based on commonly used Alex-net, network is optimized with batch normalization to improve efficiency. The experimental results indicate that the preprocessing methods increase the accuracy and decreases the time of training process of tongue shape classification, which proves that the method is effective for the recognition of different tongue shapes.