TR2022-005

Overhead Reduction for Graph-Based Point Cloud Delivery Using Non-Uniform Quantization


Abstract:

Graph-based compression can compact the signal energy of the three-dimensional (3D) point cloud and realize high- quality 3D point cloud delivery over wireless channels. However, it requires significant communication overhead of graph Fourier transform (GFT) orthogonal matrix. For significant overhead reduction, our scheme integrates two methods: Givens rotation and non-uniform quantization. The Givens rotation transforms the GFT orthogonal matrix into angle parameters. The angle parameters are then non-uniformly quantized based on an empir- ical concave cumulative distribution function (CDF). Evaluation results show the proposed scheme reduces 28.6% communication overhead and improve up to 3.8dB 3D reconstruction quality compared with the conventional schemes.