Deep Learning-based Incomplete Regions Estimation and Restoration of 3D Human Face Point Cloud

Incomplete regions in the 3D human face point cloud cause performance degradation in many applications such as human face recognition, classification, segmentation, upsampling, etc. Many methods have been studied to the completion of missing regions in different categories of datasets. However, it is a challenging task to estimate the incomplete regions in the 3D human face point cloud dataset owing to the complex shape and surface structure. In this article, we propose a novel approach to estimating and restoring 3D points in missing regions. First, we project the 3D points into a 2D depth map and use a deep learning model to estimate the missing parts. Then the complete 2D depth map is used to reconstruct the 3D point in incomplete regions. We also restore the estimated points using a deep learning model to improve the quality of the data. The proposed method assures better performance than the state-of-the-art methods in estimation and restoration of the 3D human face point cloud.