Real-time Recognition of 6 Basic Facial Expressions

This paper deals with the real-time recognition of six basic facial expressions. In order to obtain the center position of both pupils, we obtain the brightness, by using a CCD camera, along a vertical line crossing over the pupil and eyebrow as base data and calculate the cross-correlation between base data and that in the given image. We extract the position of right and left pupils separately. By using transputer, the time needed is about 40 [ms] to obtain right and left pupil's position. As the facial information for utilizing the recognition of facial expression, we use brightness data of 13 vertical lines (facial information), determined empirically and including the areas of eyes, eyebrows and mouth. Then we acquire the facial information of 6 basic facial expressions for 30 subjects whose face images have already been obtained. Since we use a layer-type neural network for recognition of facial expressions, facial information for some of 30 subjects is used for training the neural network and recognition tests done by using facial information not used for neural network learning. We find that, when we use 15 subjects for the network training, the correct recognition ratio reaches 85%, and the total time for detecting right and left pupil positions plus the recognition of facial expression is about 55 [ms] per one recognition cycle.