Learning Three-dimensional Face recognition from Sparse views for Robust Identity verification

This paper presents a deep learning-based approach to accurately reconstruct and verify 3 Dimension facial models using sparse visual inputs. Traditional 3 Dimension face recognition systems often require high-resolution or multiple-angle scans, which limit their application in real-world settings, especially with constrained devices like mobile phones or security cameras. Our method leverages sparse multi-view inputs—typically as few as two or three images—to predict complete 3 Dimension geometry and extract robust identity features. We propose a hybrid model that integrates point cloud-based learning with facial landmark-guided refinement for accurate representation and matching. This approach not only reduces the dependency on expensive capture hardware but also enhances robustness against occlusions, varied lighting, and facial expressions. Experimental results on standard datasets show that our model maintains high verification accuracy even under extreme sparsity. Furthermore, it demonstrates superior generalization in low-data scenarios. This work has promising applications in biometric authentication, digital forensics, and virtual avatar creation.