Human facial expression recognition using wavelet transform and hidden Markov model

The accuracy of the Facial Expression Recognition (FER) system is completely reliant on the extraction of the informative features. In this work, a new feature extraction method is proposed that has the capability to extract the most prominent features from the human face. The proposed technique has been tested and validated in order to achieve the best accuracy for FER systems. There are some regions in the face that have much contribution in achieving the best accuracy. Therefore, in this work, the human face is divided into number of regions and in each region the movement of pixels have been traced. For this purpose, one of the wavelet families named symlet wavelet is used and individual facial frame is decomposed up to 2 levels. In each decomposition level, the distances between the pixels is found by using the distance formula and by this way some of the informative coefficients are extracted and hence the feature vector has been created. Moreover, the dimension of the feature space is reduced by employing a well-known statistical technique such as Linear Discriminant Analysis (LDA). Finally, Hidden Markov Model (HMM) is exploited for training and testing the system in order to label the expressions. The proposed FER system has been tested and validated on Cohn-Kanade dataset. The resulting recognition accuracy of 94% illustrates the success of employing the proposed technique for FER.