PLGP: point cloud inpainting by patch-based local geometric propagating

The booming of LiDAR technologies has made the point cloud become a prevailing data format for 3D object representation. However, point cloud usually exhibits holes of data loss mainly due to occurrence of noise, occlusion or the surface material of the object, which is a serious problem affects the target expression of point cloud. Point cloud inpainting is the key solution for holes problem. In this paper, we propose a patch-based local geometric propagating (PLGP) method to automatically fill the lost data obtained by three-dimensional scanning. Different from typical methods transforming the point cloud into range image to conduct the hole-detection or filling the missing region with a whole best match, this work tends to detect the hole directly in 3D space and inpaint it by iteratively searching for the context with local similarity and making it propagate appropriately along the occlusion’s local geometric structure. The experimental results with comparisons demonstrate its competitive effectiveness with a F-score as high as 0.89 and a 23.45 dB average gain in GPSNR with consumed time reduced by up to 60%.