Information Fusion Techniques Applied In Traditional Chinese Tongue Diagnosis System

Tongue diagnosis is one of the most valuable diagnostic methods of TCM. In recent years, as the rapid development of computer science, orienting to computerization has already become the absolute trend for traditional TCM tongue diagnosis. The purpose of this dissertation is contributing to the computerization of TCM tongue diagnosis by introducing information fusion theory to TCM tongue diagnosis research.The major contributions of this dissertation are: designing the TCM diagnosis system on the basis of Information Fusion theory; proposing and realizing self-adaptive glisten points detection and filling algorithm based on Data Layer Fusion; with Feature Layer Fusion theory, associating the color, texture, shape, moist and dry features of the tongue together into a single feature vector, and applying it to classify TCM diseases; designing and realizing the Decision Layer Fusion model, forming a TCM tongue diagnosis system based on combining classifiers. Through evidence of experiment, the accuracy of the system was remarkably raised.First, in the application of Data Layer Fusion, this dissertation studies glisten points detection algorithm in the pre-processing period and proposes a new algorithm based on Data Layer Fusion theory, which gains well self-adaptation; after that, applies a value-inserting algorithm in common use of Data Layer Fusion to fill the glisten points on the tongue surface, which is propitious to the processing afterwards.Second, in the application of Feature Layer Fusion, this dissertation exacts the moist/dry features and shape of the tongue, and picks up the color and texture features in multiple feature space, forming the associated feature vector. After that, PCA method and Boosting Feature Filtering are used respectively to descend the dimensions of the vector, which is employed in the following training of classification.Third, in the application of Decision Layer Fusion, the Bayesian Network method is utilized to train a model with probability outputs, and the Adaboost algorithm in re-sampling technique is applied with Bayesian Network to get multiple classifies. Afterward, the majority voting method is used to form a combined classifier, whose accuracy is remarkably higher than single classifier.Last, the three layers of fusion are combined together to shape a primary design of TCM tongue diagnosis system based on Information Fusion model, which is an instructive and groping trial in associating TCM tongue diagnosis and Information Fusion theory.