Recognition of Mixed Facial Expressions by Neural Network.

Abstract

Since the facial expressions of human emotions are usually not distinct, the development of a technique for machine recognition of mixed facial expressions is very crucial for the true realization of AHI (active human interface) between machine and human beings. This paper deals with a neural network method for machine recognition by decomposing mixed facial expressions into 2 or 3 components of 6 basic facial expressions. We obtain the images of mixed facial expressions from video tape which recorded continuous change in facial expressions, and using the information of the (x, y) coordinates of facial characteristic points (FCP) for 19 subjects, the neural network is trained. The recognition test is conducted by inputting facial information not used in training the neural network to the trained neural network. The neural network method is found to give a rather high agreement rate of about 70% compared with recognition results obtained by human beings.