Master Thesis Title: Facial Expersion recognition using soft computing, 2014
Fuzzy Inference systems have been successfully applied for different pattern recognition problems. However, in some real-world applications such as facial expression recognition, the uncertainties related to the feature space may be so high that it is hard to model the feature space by fuzzy membership functions. On the other hand, it is believed that type-2 fuzzy sets have high potentials of uncertainty management in the space of features. However, adjusting the parameters of Type 2 membership functions is a difficult task. In this paper, we analyze the effect of incorporating type-2 fuzzy system into facial expression recognition problem. In this regard, two adaptive type-2 fuzzy models are proposed. The first model employs interval type-2 Mamdani fuzzy system for constructing fuzzy face space and uses Genetic Algorithm for optimizing the membership functions parameters. The second model is an interval type-2 Neuro-Fuzzy system which contains interval type-2 fuzzy sets as membership functions. In this case, the gradient descent algorithm is utilized for tuning the parameters of this system. Numerical results demonstrate the superiority of type-2 fuzzy systems with respect to the corresponding type 1 systems and show that the proposed systems can better cope with the uncertainties in facial expression recognition.