Projection Difference-Guided Geometry Quality Enhancement for Video-Based Point Cloud Compression

Video-based point cloud compression (V-PCC) developed by MPEG has achieved remarkable compression efficiency for dynamic point clouds. However, the point clouds compressed by V-PCC still suffer from serious artifacts due to the lossy compression and lose a large number of points. In this paper, we propose a new geometry quality enhancement method for the V-PCC compressed point clouds and it can effectively recover the lost points. Our method is applied to the 2D projected near and far frames rather than 3D point clouds. It is designed to enhance the quality of 2D frames, guided by the predicted difference information between them. More specifically, we firstly construct a gradient-based difference prediction network (G-DPnet) to predict the difference between near and far frames. This difference is introduced in the enhancement of 2D frames, for the recovery of the lost 3D points. Meanwhile, we propose the single-frame quality enhancement network (SFQEnet) to separately enhance near and far frames. The enhanced frames are then used to produce the near-far frame difference with G-DPnet. After obtaining the difference, we feed it into a dual-frame quality enhancement network (DFQEnet) to guide the further enhancement of near and far frames. Experimental results demonstrate that our method can effectively recover a large number of lost points and improve the quality of point clouds compressed by V-PCC.