Measurement of the Strength of Six Basic Facial Expressions by Neural Network.

Abstract

The object of this study is to develop an "Active Human Interface (AHI)" that realizes communication of emotion between machines (computer and/or robot) and human beings. In establishing the new paradigm for the realization of AHI, as a first step, we have investigated the method of machine recognition of human emotion. This paper investigates the method of measuring the strength of six basic emotions rom facial expressions by means of a neural network. From continuous videotape recordings of subject's facial images, we select a weak facial expression for each of the 6 basic emotions from 19 subjects. We obtain the strength for each of these weak facial expressions by interviewing these clients under the condition that the strength of the basic facial expressions used above is equal to l and normal facial expression is equal to 0. The results are used for training the neural network for each of the basic facial expressions. The trained neural network is tested with facial image data other than those used in training, and the neural network method is found to be effective in measuring the strength of facial expressions since the method shows a rather high agreement with the recognition results of human beings.