Human Model Reconstruction with Facial Region Optimization: A Fusion Nerf Approach
Qiuhong Zheng, Jinghan Wang, Yun Shen · 2024
The existing parameterized model-based 3D human reconstruction methods typically does a global representation of the body, lacking further optimization and design for the facial region, leading to poor representation in facial region during practical use. However, a qualified human model demands detailed facial modeling to fully convey facial information, thus avoiding issues like unclear facial identity expression and inadequate model refinement. This paper proposes a 3D human model reconstruction method based on the fusion of face and body using neural radiation field (NeRF). NeRF is established separately for the 2D image’s face and body parts, followed by fusion to reconstruct a finely detailed facial model. Additionally, a texture resampling model correction mechanism based on RePaint is proposed to ensure the coherence of the body structure and uniformity of texture colors in the fusion region of face and body. Ultimately, a 3D human model with a refined facial region is recon-structed.