This research develops a system which can recognize three-dimensional human facial expression. In this study, Dynamic Skin Color Scale (DSCS) and Grey Edge Method are used to capture the key feature points. DSCS techniques have the advantage of less effect by light source. In order to effectively obtain the feature point and minimize the error comparison, the recognition process is divided into two steps and combined with RGB contrast enhancement. In addition, some affine coefficient invariable theorem are used as the basis for recognition. The selection of system reference plan is at the eye position, therefore, the feature point will not missing when the face is moving. In order to enhance the robust recognition, many types of expression training are added to each Relative Affine coefficient.