中文摘要：

针对在Candide-3模型上进行三维人脸重建时不能表达人眼部细节动作的问题，提出了一种改进Candide-3模型重建人脸的方法。该方法首先在Candide-3模型上对眼部进行建模，其中眼部的建模具体分为眼球的建模以及眼睑的建模。对于眼球采用2个半球结合的方式进行建模。对于眼睑的建立，分别在Candide-3的模型上对眼睑的内边缘、外边缘以及上眼睑沟用Bezier曲线进行建模。然后结合人脸特征点的信息和局部最小二乘法匹配原理建立了与输入人脸图像相仿的三维人脸模型。接着，对脸部进行特征点约束的方法贴纹理，眼球分别用局部球形纹理贴图以及球形纹理贴图的方法贴图。最后，在该模型上用分段线性插值的方法，逼真地完成了眼睑的睁闭、眨眼以及眼球的旋转等相关动作。

英文摘要：

A face reconstruction approach based on the improved Candide-3 model is presented in order to solve the problems that the details of eye movements cannot be expressed when reconstructing a three-dimensional facial mod-el in Candide-3 model.First, the eye model was built in the Candied-3 model.It was specifically divided into two parts：the eyeball modeling and the eyelid modeling.The eyeball modeling was conducted in combination with two hemispheres.For the modeling of the eyelid, the Bezier curve was used to model the inner and outer edges of eyelid and the upper lid groove in the Candide-3 model.Next, by combining the information of facial feature points and lo-cal least squares matching principle, the 3D face model similar to the input face image was built.After that, paste facial texture by feature point constraint methods and paste texture to eyeball by the spherical and partially spherical texture mapping methods, respectively.Finally, some actions, such as closing and opening of eyelid, blinking, ro-tation of eyeball and other related actions were realistically completed by the piecewise linear interpolation method in the built model.