Abstract:

This paper studies the Isomapbased nonlinear dimensionality reduction and
visualization method,applies it to the feature reduction of facial animation parameters of
emotional facial image sequences,and analyzes the relationship between the reduced
dimension space and the cognitive psychological emotion space.The experimental results show
that the coordinate of the manifold feature in the reduceddimension space shows the
emotion intensity of the corresponding image frame,and the intensity curve of an emotional
facial sequence from Isomap is more reasonable and smooth than that from PCA. The visual
emotion recognition results also show that the mainfold features from Isomap have better
performance than the original features and the manifold features from PCA.Moreover,the
recognition rates are more balanced between different emotions.

Key words: Isomap;manifold;emotional space;nonlinear feature reduction;emotional intensity