Influence-Based Quantization Method for Lifting Transform of G-PCC

Abstract

Point cloud is one of the emerging 3D representations used in diverse applications. However, a point cloud typically comprises a huge amount of data; thus, compression is crucial for efficiently transmitting and storing. To meet these requirements, moving picture experts group (MPEG) developed geometry-based point cloud compression (G-PCC). In G-PCC, lifting transform is used as one of the attribute coding methods and it relies on level-of-detail (LoD) structure for transform. The LoD-based lifting transform makes each point have different influence on the transformation. To utilize these characteristics to quantization process, this paper proposes an influence-based point-wise quantization method for the lifting transform of G-PCC. The proposed quantization method includes two novel techniques: influence-based point-wise scaling (IPWS) and influence-based rate-distortion optimized quantization (IRDOQ). The IPWS is a novel scaling function that improves the existing adaptive scaling method. The IRDOQ determines the optimal levels by balancing the rate and distortion of each lifting coefficient based on the point-wise influence weight. The influence weight indicates the frequency with which its current point is referenced for prediction in the lifting transform. The influence weights calculated during the lifting transform are reused for the proposed IPWS and IRDOQ. The proposed method enhances the overall coding performance by adjusting the balance between rate and distortion based on the influence of each lifting coefficient on the quantization process. The experimental results demonstrate that the proposed method achieves an average Bjøntegaard-Delta rate (BD-rate) improvement of 2.0%, 5.5%, 5.4%, and 3.6% for the luma, chroma Cb, chroma Cr, and reflectance, respectively, when compared to the G-PCC reference software.

Keywords

• Immersive media
• G-PCC
• point cloud compression
• RDOQ