Genetic Algorithm and Neural Network for Face Emotion Recognition

Abstract

Human being possesses an ability of communication through facial emotions in day to day interactions with others. Some study in perceiving facial emotions has fascinated the human computer interaction environments. In recent years, there has been a growing interest in improving all aspects of interaction between humans and computers especially in the area of human emotion recognition by observing facial expressions. The universally accepted categories of emotion, as applied in human computer interaction are: Sad, Anger, Joy, Fear, Disgust (or Dislike) and Surprise. Ekman and Friesen developed the most comprehensive system for synthesizing facial expression based on what they called as action units (Li, 2001). In the early 1990’s the engineering community started to use these results to construct automatic methods of recognizing emotion from facial expression in still and video images (Sebe, 2002). Double structured neural network has been applied in the methods of face detection and emotion extraction. In this, two methods are proposed and carried out; they are lip detection neural network and skin distinction neural network (Takimoto et al., 2003). Facial action coding (Panti & Patras, 2006) is given to every facial points. For example, code 23 is given for lip funnel, code 4 for eye brow lower, code 10 for chin raise etc. The cods are grouped for a specific facial emotion. In order to determine the category of emotion, a set of 15 facial points in a face-profile sequence has been recommended. The work performs both automatic segmentation of an input video image of facial expressions and recognition of 27 Action Units (AUs) fitted to facial points. A recognition rate of 87% has been reported. The motion signatures (Anderson & Peter, 2006) are derived and classified using support vector machines (SVM) as either non-expressive (neutral) or as one of the six basic emotions. The completed system demonstrates in two simple but effective computing applications that respond in real-time to the facial expressions of the user. The method uses edge counting and image correlation optical flow techniques to calculate the local motion vectors of facial feature (Liyanage & Suen, 2003). Cauchy Naive Bayes classifier is introduced in classifying the face emotion. The persondependent and person-independent experiments have demonstrated that the Cauchy distribution assumption typically provides better results than those of the Gaussian distribution assumption (Sebe, 2002).