Fig. 1. Flowchart of 3D face reconstruction

Fig. 2. Results of reflected light separation. (a) Vertically polarized image; (b) parallel polarized image; (c) specular reflection image

Fig. 3. Gradient light images in different directions. (a) Along the X direction; (b) along the -X direction; (c) along the Y direction; (d) along the -Y direction; (e) along the Z direction; (f) all light

Fig. 4. Normal mapping principle

Fig. 5. Image acquisition device. (a) Shooting device; (b) relative position of the cameras

Fig. 6. Experiment of specular reflection and diffuse reflection separation

Fig. 7. Results of normal map fusion. (a) Diffuse reflection normal map; (b) specular reflection normal map; (c) fusion normal map

Fig. 8. Comparison of normal map reconstruction with different fusion modes. (a) Low precision model; (b) reconstruction with 2∶1 fusion ratio; (c) reconstruction with 5×5 high-pass filtering and 1∶1 fusion ratio; (d) reconstruction with 5×5 high-pass filtering and 2∶1 fusion ratio; (e) reconstruction with 7×7 high-pass filtering and 1∶1 fusion ratio; (f) reconstruction with 7×7 high-pass filtering and 2∶1 fusion ratio

Fig. 9. Multi-view images. (a) Multi-view images; (b) multi-view normal maps

Fig. 10. Results of fine 3D face reconstruction

Fig. 11. Comparison of the render of highprecision model and normal map model. (a) Model1; (b) model2; (c) model3; (d) rendering of model1; (e) rendering of model2; (f) rendering of model3