Fig. 1. Flow chart of our method
Fig. 2. Positive face feature points extracted by ASM. (a) Positive face 1; (b) positive face 2
Fig. 3. Feature points extracted by different methods. (a) ASM; (b) PRN
Fig. 4. Facial edge features extracted based on Canny operator. (a) Face image; (b) face edge
Fig. 5. Superposition of face edge and face feature points. (a) Face edge; (b) face feature points;(c) superimposed image
Fig. 6. Structure of the edge feature detector
Fig. 7. Feature points before and after calibration. (a) Before calibration; (b) after calibration
Fig. 8. Projection results. (a) 3D face model; (b) projected image
Fig. 9. Result of model rendering. (a) Before rendered; (b) after rendered
Fig. 10. NMSE with different coefficients. (a) λid; (b) λexp
Fig. 11. NMSE of reconstructed image by different methods
Fig. 12. 3D face reconstruction results. (a) 2D image; (b) 0°; (c) 30°; (d) 60°; (e) 90°
Fig. 13. Reconstruction results of different methods. (a) 2D image; (b) VRN method; (c) 3DDFA method; (d) 3DMM-CNN method; (e) our method
Method | VRN | 3DDFA | 3DMM-CNN | Ours |
---|
AFW | 0.698 | 0.634 | 1.223 | 0.617 | IBUG | 0.623 | 0.644 | 1.056 | 0.545 | LFPW | 0.669 | 0.653 | 1.326 | 0.656 |
|
Table 1. NMSE of reconstructed image by different methods