Cross-Attribute Prediction for Enhanced Geometry-Based Point Cloud Compression

Three-dimensional (3D) point cloud is a representative media that can be used in many applications such as metaverse and immersive communication. Under limited storage and bandwidth conditions, efficient point cloud compression (PCC) play a crucial role. The Moving Picture Experts Group (MPEG) is actively developing the geometry-based point cloud compression (G-PCC) standard and plan to release the second edition of G-PCC, namely Enhanced G-PCC. However, when encoding point clouds with fused attributes (e.g., color, reflectance), redundancy between the two attributes remains unaddressed. We investigate the correlation between the two attributes of level of detail-based consecutive points in predictive transform (PT) and octree layer-based consecutive blocks in region-adaptive hierarchical transform (RAHT), and propose a cross-attribute prediction (CAP) method. In PT, neighbor points are used to fit the correlation between the two attributes, and then the reconstructed first attribute is used to predict the second attribute. In RAHT, the difference between the first attribute of the current block and the candidate prediction block is used to reselect the intra-prediction eligibility of the candidate blocks for the second attribute. Experimental results show that the coding efficiency of the proposed method overwhelms that of the state-of-the-art Enhanced G-PCC reference software. Specifically, by following the common test condition, CAP can achieve average BD rates of -0.12%, -0.65%, -0.17%, and -3.56%, for Luma, Cr, Cb and Reflectance under the C2 condition (lossy geometry with lossy attributes) in PT and -0.69%, -0.42%, -0.33%, and -1.19%, under the C2 condition in RAHT, compared to TMC13 version 28.0. It should be noted that BD rates of -1.41% and -2.25% could be achieved when using 1-PCQM to evaluate the perceptual quality under the above two conditions.