Facial Expression Recognition Based on Emotion Dimensions on Manifold Learning

Abstract

This paper presents a new approach method to recognize facial expressions in various internal states using manifold learning (ML). The manifold learning of facial expressions reflects the local features of facial deformations such as concavities and protrusions. We developed a representation of facial expression images based on manifold learning for feature extraction of facial expressions. First, we propose a zero-phase whitening step for illumination-invariant images. Second, facial expression representation from locally linear embedding (LLE) was developed. Finally, classification of facial expressions in emotion dimensions was generated on two dimensional structure of emotion with pleasure/displeasure dimension and arousal/sleep dimension. The proposed system maps facial expressions in various internal states into the embedding space described by LLE. We explore locally linear embedding space as a facial expression space in continuous dimension of emotion.KeywordsFacial ExpressionInternal StateFace ImageEmotion WordFacial Expression RecognitionThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.