Automatic extraction of facial organs and recognition of facial expressions

This paper deals with the method to realize automatic contour extraction of facial organs such as eyebrows, eyes and mouth for the deformed face, and automatic categorization and recognition of facial expressions by using unsupervised neural network. We define the elastic contour model in order to hold the contour shape and then determine the elastic energy acquired by the amount of modification of the elastic contour model. We also define the image energy obtained by brightness differences of the control points on the elastic contour model. Applying the dynamic programming method, we determine the contour position where the total value of the elastic energy and the image energy becomes minimum. We use the transformation value of the control points on the elastic contour model as a facial information for the neural net training and for the recognition test obtained from 20 subjects in terms of 6 typical facial expressions. We found that a 76% correct recognition rate was achieved.