The purpose of this paper is to estimate the degree of human facial expression from expressionless to its maximum. For the purpose of extracting subtle changes in the face, it is necessary to eliminate the individuality appearing in the facial images. Our method is based on the idea that the degree of facial expression can be extracted as a variation from an expressionless face. Using a potential net, a face is sampled as a whole pattern from a facial edge image which we regard as a potential field. Then applying the Karhunen-Loeve expansion, the emotion space is achieved with principal components, and estimation is achieved by projecting input images onto the eigenspace. We have constructed three kind of expression models: happiness, anger, and surprise. The input images are evaluated.
展开▼
