Precise surface normal estimate is typically necessary for photometric stereo, an essential technique for 3D reconstruction. Singular Value Decomposition (SVD) is still a popular method, however the quality of the reconstruction may suffer from its flaws. This paper proposes a new approach to enhance initial normal vector estimations from SVD using Particle Swarm Optimization (PSO). We obtain an optimal solution close to the SVD result by utilizing PSO’s powerful exploration capabilities. This enhanced estimator far exceeds the constraints of traditional SVD-based approaches in terms of accuracy and resilience. Our method opens the door to better 3D reconstruction with increased dependability and authenticity.