Emotion Recognition using DWT, KL Transform and Neural Network

Human face communicates important information about a person’s emotional condition. In this paper an approach for facial expression recognition using wavelet transform for feature extraction and neural network classifier for five basic emotions is proposed. The strength of the algorithm is the reduction in feature size and use of less number of images for training the network, compared to existing similar approaches. Static images of the Cohn-Kanade Face Expression image database have been used for experimentation. The facial expression information that are mostly concentrated on mouth, eye and eyebrow regions are segmented from the face. Then the low-dimension features are acquired using 2-level Discrete Wavelet Transform and Karhunen–Loeve transform. A neural network classifier is constructed to categorize the emotions. The neural network based classifier yielded an average accuracy of 96.4%. The expressions that are recognized are happiness, sadness, anger, surprise and disgust.