Abstract
This paper describes the extraction of psychometrical characteristics of face pattern used in Faces Method-a method of representing multidimensional data by facial expressions-and their application in representing system states.
First, the configuration of similarity of the 88 faces constructed by uniform random numbers based on subjects' judgment is obtained by multidimensional scaling and the psychometrical distances between the 88 faces in the configuration are calculated. The 88 faces are subsequently classified into 15 groups by using the distances and for each group a representative facial expression is extracted. The language to interpret the classified 15 faces is examined by using 16 words expressing varying forms of emotion, e.g., ‘surprise’, ‘anger’, ‘happiness’, etc. By the method of equal appearing intervals, one dimensional scaling of the 88 faces is also obtained according to the subjects' judgment in the experiment of grading the facial pleasantness and unpleasantness. It can be inferred therefore from this experiment that the method can be used in data evaluation.
A technique of assigning data characteristics to typical facial expressions is developed using the above results. This technique is primarily to determine a mapping function of the data space to the facial expression space by GMDH. The representation of the dynamic data of one heat exchanger plant model is taken as a practical illustration of the technique. The transition of the plant from a stationary state to an abnormal state is depicted by using facial expressions showing pleasantness, normal, surprise, and anger.
