ANN based human facial expression recognition in color images

This paper describes an ANN based human facial expression recognition method. An automatic technique is generated for detecting 22 most important facial feature point and generation of facial feature vector by finding the Euclidian distances between some particular points. Here the key facial regions are found based on a defined face model. The face model is generated by the detection of the eyes and mouth points. Skin color detection is done by 2D Color Space Skin Clustering Method and face detection is done by connected component analysis followed by a set of face heuristics. Feed forward back Propagation neural network is used as the network classifier to classify the facial expression from a set of seven basic expressions like happy, sad, surprise, fear, anger, disgust and neutral. The experiment is done on Color FERET Database and got an accuracy of 100% for trained dataset and 85% accuracy for test set.