Semantic knowledge based on fuzzy system for describing facial expression

We propose a new approach to describe facial expressions based on axiomatic fuzzy set (AFS), which can convert the features to semantic knowledge. The method has two advantages: the first one is that the semantic concepts are gained by data distribution. It means that the concepts can be directly obtained from data set, and the priori model is not required. The second one is the semantic concepts can be operated for representing expression characteristics. Then, an optimized criterion is designed to select the best semantic concept set for representing each expression. Finally, we execute the experiments on CK+ database, and have an interpretation for corresponding expression. In addition, the performance of this method is estimated with state-of-the-art methods such as C4.5, Repeated Incremental Pruning to Produce Error Reduction(Ripper), DecisionTable, Cart.