基于三角坐标系的人脸表情方向性研究

Research on the Directionality of Facial Expression Based on Triangle Coordinate

为了提取准确的人脸表情特征,采用三角坐标系对人脸表情进行快速归一化,提出人脸表情变形矢量坐标系,并验证了人脸表情特征的方向性.首先给出三角坐标系的定义;然后建立人脸特征点为基础的三角网格坐标系,利用轮廓三角形进行归一化处理后得到各个特征点坐标位移,映射到高维空间建立人脸表情位移矢量坐标系.通过不同人同一种表情的移植、同一个人同种表情的渐变和分层方向基神经网络对人脸表情识别这3个实验,证明了三角坐标系下人脸表情特征在高维空间呈现方向性分布.

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 isobtained, which is then mapped to high dimensional space to create a facial expression displacement vectorspace. Finally, the directionality of facial expression features existing in the high dimensional space isproved 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.