Lossy Point Cloud Attribute Compression with Subnode-Based Prediction

Abstract: Recent years have witnessed that 3D point cloud compression (PCC) has become a research hotspot both in academia and industry. Especially in industry, the Moving Picture Expert Group (MPEG) has actively initiated the development of PCC standards. One of the adopted frameworks called geometry-based PCC (G-PCC) follows the architecture of coding geometry first and then coding attributes, where the region adaptive hierarchical transform (RAHT) method is introduced for the lossy attribute compression. The upsampled transform domain prediction in RAHT does not sufficiently explore the attribute correlations between neighbor nodes and thus fails to further reduce the attribute redundancy between neighbor nodes. In this paper, we propose a subnode-based prediction method, where the spatial position relationship between neighbor nodes is fully considered and prediction precision is further promoted. We utilize some already-encoded neighbor nodes to facilitate the upsampled transform domain prediction in RAHT by means of a weighted average strategy. Experimental results have illustrated that our proposed attribute compression method shows better rate-distortion (R-D) performance than the latest MPEG G-PCC (both on reference software TMC13-v22.0 and GeS-TM-v2.0).

Keywords: point cloud compression; MPEG G-PCC; RAHT; subnode-based prediction