Table 1 Comparison of the previous methods and the proposed method
From: Single view-based 3D face reconstruction robust to self-occlusion
Categories | Methods | Strengths | Weaknesses |
|---|---|---|---|
Frontal view | - No requirement for training data | - Infeasible constraints (Lambertian reflectance model and known light source direction ) | |
|  | Modeling the relationship between intensities and depths [13–16] | - Low computational complexities | - Requires pixel-by-pixel alignment between intensities and depth map - Requirement of training data |
| Â | - Low computational complexity | - Requirement of training data | |
Arbitrary view | - Robust to pose and illumination variation - Less sensitive to self-occlusion | - High computational complexity - Dense correspondence between faces - Requirement of training data | |
| Â | Simplified version of 3D morphable model - EM algorithm [20] | - Low computational complexity | - Requirement of training data - Self-occlusion problem |
| Â | Simplified version of 3D morphable model - Proposed method | - Low computational complexity - Robust to self-occlusion | - Requirement of training data |