Abstract: The synthesis of human face is difficult for real-time interaction application, because it is affected by the light, background, gesture of head, etc. In order to generate the synthesis image of human face for virtual display, a quick sample-based learning method is proposed to detect human face from camera image before the display desk, then through the parameterized model of human face samples, certain feature points on human faces are tracked. Finally the synthesis images of perfect faces for different human faces are generated in real time, which is successfully used in virtual display of science and technology museum.

Key words: face synthesis, virtual display, perfect face, feature extraction