A method for constructing 3D human facial images which maintains the characteristics of video input image is proposed. First, a personalized 3D polygonal facial model is constructed from two orthogonal facial images. Video input frames are modified, and the vertices composing the 3D polygons are altered according to the contraction rule of facial muscles. Moreover, the modified input frame is texture mapped directly onto to the forementioned model to improve the quality of the reconstructed image. This method was implemented onto an ordinary PC, and an experiment using five persons' face data showed the effectiveness of the method both quantitatively and subjectively.