Face expression recognition means analyzing specified face expression and its’ change through computer technology, and realizing more natural and intelligent man-machine supervision. The purpose of the research of face expression recognition is to recognize people’s expression automatically by AI products. This research is a very challenging project which involves pattern identification、machine vision、motion tracking、and psychology and many other technologies in various fields.Unascertained cluster is an important method of multivariate statistical analysis and important tool of pattern identification; it has great impact on auto control, system identification, fault diagnosis and many other fields. Because the classification and description of face expressions and people’s emotions have a close association, it has constant uncertainty. And this problem could be solved properly by taking the advantage of unascertained cluster to commence the classification of expressions and to make it more circumstantial and objective. This article quotes the analysis of unascertained cluster in the section of classifying Face expressions.At first, this article summarizes the background of this research, and analyzes existing methods for the feature extraction and recognition of face expression and, on the basis of it, presents the face recognition technology based on analysis of unascertained cluster. This project mainly includes:(1) Characterizing a model of the feature of expressions in order to extract effective feature of expressions by extracting shape information and texture information through the fixed points of face image on the utilization of subjective motility model, and reducing the dimensions of the feature vectors with the analysis of principle component in the process of using the subjective motility model.(2) Leading unascertained cluster into face expression recognition according to the uncertainty of face expressions, and commencing the training of features for the feature model to get the temples to characterize each expression, and determining the type of the expression of sample to be measured with the nearest neighbor classifier. |