Abstract: To extract features of facial expression accurately,this paper proposes a kind of facial expression displacement vector space based on normalized expressions within triangle coordinate.Firstly,the definition of the triangle coordinate is given.Then a triangle coordinate grid system is built based on points of facial features.After normalizing contour triangle in the system,coordinate displacement of every feature point is obtained,which is then mapped to high dimensional space to create a facial expression displacement vector space.Finally,the directionality of facial expression features existing in the high dimensional space is proved by performing three experiments:the transplant of same facial expression type of different individuals,the slow transition of same human facial expression type of same individual,and facial expression recognition using DBF neural network.