Unsupervised Temporal Segmentation of Facial Behaviour

In this work, we attempt to tackle the problem of automatic facial expression recognition using unsupervised clustering algorithms. Our features consist of Active Appearance Model tracked facial interest points. We explore the use of various clustering algorithms like k-means, Gaussian Mixture Models and a novel temporal clustering algorithm Aligned Cluster Analysis [7]. The Extended Cohn-Kanade database [6] is used for testing the algorithms and the results indicate the need for better clustering algorithms and better choice of feature vectors.