Fig. 1. Schematic of sampling point distribution. (a) Standard rectangular distribution; (b) non-standard rectangular distribution

Fig. 2. Schematic of filling non-matrix data

Fig. 3. Height map corresponding to three distributions of sampling points. (a) Standard rectangular distribution; (b) barrel distribution; (c) pillow distribution

Fig. 4. Surfaces reconstructed by proposed method. (a) Standard rectangular distribution; (b) barrel distribution; (c) pillow distribution

Fig. 5. Reconstruction error corresponding to standard rectangular distribution. (a) Proposed method; (b) Southwell method; (c) LSI-T method

Fig. 6. Reconstruction error corresponding to barrel distribution. (a) Proposed method; (b) Southwell method after resampling; (c) LSI-T method

Fig. 7. Reconstruction error corresponding to pillow distribution. (a) Proposed method; (b) Southwell method after resampling; (c) LSI-T method

Fig. 8. Schematic of the position of the target and blank points in the matrix. (a) Circular area; (b) area with small holes

Fig. 9. Comparison of computing time in two areas

Fig. 10. Influence of the number of sampling points on the computing time

Fig. 11. Setup of polarization reconstruction method based on circular polarized light

Fig. 12. Intensity at different rotation angles of the wave plate within the target area

Fig. 13. Gradient distribution in the target area. (a) Gradient along x direction; (b) gradient along y direction

Fig. 14. Reconstructed surface in the experiment

Fig. 15. Reconstruction error map by different methods. (a) Proposed method; (b) Southwell method after resampling