A Novel LDA and HMM-Based Technique for Emotion Recognition from Facial Expressions

Automatic human emotion recognition from facial expressions has grabbed the attention of computer scientists for long, as it is necessary to understand the underlying emotions for effective communication. Over the last decade, many researchers have done a lot of work on emotion recognition from facial expressions using the techniques of image processing and computer vision. In this paper we explore the application of Latent Dirichlet Allocation (LDA), a technique conventionally used in Natural text processing, used with Hidden Markov Model (HMM), for learning the dynamics of facial expression for different emotions, for the classification task. The technique proposed works with facial image sequence. Each frame of an image sequence is represented by a feature vector, which is mapped to one of the words from the dictionary generated using k-means. Latent Dirichlet Allocation then models each image sequence as a set of topics, where each topic is in turn probability distribution over words. LDA doesn‟t take into account the order of words which appear in an image sequence to find topics. However we have the information of the order in which the words and hence the topics appear in an image sequence. We leverage the information about the order of topics for classification in the next step. This is done using a separate Hidden Markov Model for each emotion, each of which is trained to recognize the dynamics of facial expressions for the emotion. The emotions dealt with are six basic emotions: happy, fear, sad, surprise, angry, disgust and contempt. We further compare our results with the technique in which sequence of words instead of topics is used by HMM for learning facial expression dynamics. The results have been presented on Extended Cohn-Kanade database. The accuracy obtained on the proposed technique is 80.77% .The use of word sequence in general performs better than use of topic sequence for learning facial expression dynamics.