Adaptive Local Reconstruction for Geometry-based Point Cloud Compression

In recent years, point cloud compression has emerged as a focal research topic. Geometry-based Point Cloud Compression (G-PCC) developed by the Moving Picture Experts Group (MPEG) has demonstrated to achieve remarkable compression efficiency across a diverse range of datasets. Triangle soup (Trisoup) is an efficient lossy geometry compression tool in G-PCC. However, its sampling rate in the geometry reconstruction stage is globally fixed, resulting in a relatively uniform distribution of the reconstructed point cloud, which cannot accurately represent the variations in density across different regions. In this paper, we propose an adaptive local reconstruction method to enhance the local quality of the point cloud reconstructed by Trisoup. On the encoding side, we first analyze the distribution density of the input point cloud in each leaf node. Based on this information, we adjust the local sampling rate of ray tracing during the geometry reconstruction process, thus generating region-specific local representations that closely resemble the ground truth. Experimental results show that the proposed method achieves an average geometry coding gain of 10.7% under D2 quality metric, along with a reduction in time complexity on both the encoder and decoder. Benefiting from the accurately reconstructed geometry, the attribute coding performance is improved by 2.1%, 2.0%, 2.5% in Luma, Chroma Cb, Chroma Cr channels, respectively.