Identifying Emotion in Static Face Images

Abstract

We use combinations of feedforward networks trained to recognize emotions in face images to achieve excellent generalization. Networks trained with an input encoding of face features (eyes and mouth) achieved about an 84% generalization rate on novel faces. A similar encoding technique applied to the entire face with the same number of parameters achieved only a 60% generalization rate. This suggests that the actual representational scheme used by the brain to identify emotions may consist of face features rather than the entire face. 1 Introduction In an extension of Cottrell and Metcalfe's work on recognizing emotions in face images [5], the performance of artificial neural networks in classification of emotions in face images is explored. In their work, undergraduates were asked to exhibit a number of different emotions. The images were then compressed with an auto-associative network, and the hidden unit activations for each image were then used as input to another network. The ou...