Visual Recognition of Emotional States

Recognizing and interpreting a human’s facial expressions and thereby his mood are an important challenge for computer vision. In this paper, we will show that trajectories in eigenspace can be used to automate the recognition of facial expressions. Prerequisite is the exact knowledge of position and size of the face within a sequence of video images. Precision and stability are two properties deciding if a tracking algorithm is suitable for subsequent recognition tasks. We will describe in this paper a robust face tracking algorithm that can sufficiently normalize a video stream to a face allowing for facial expression recognition based on eigenspace techniques. The presented face tracking algorithm is mainly based on using probability maps generated from color histograms.