Challenge
A 3D point cloud represents a set of points embedded in 3D space and plots each point’s coordinates along with the attribute information associated (i.e., photometric properties). Different from the conventional video techniques, 3D point cloud systems are more effective and efficient in representing objects and scenes of complex topology, which is enabled by free-viewpoint rendering. Despite the merits, 3D point clouds require enormous amounts of storage and huge bandwidths for transmission, as they contain millions up to even billions of 3D points carrying the geometry and attribute information. Furthermore, it is more challenging to compress 3D point clouds compared to traditional image/video compression, as 3D points are distributed randomly in 3D space leading to an irregular structure. This project will therefore explore efficient compression schemes to significantly improve the coding efficiency of the 3D point cloud compression system.
Focus
(1) To establish the metric for subjective visual quality assessment of 3D point clouds and to study the Rate-Distortion Optimization (RDO)-based encoding methodology accordingly. This study provides both the subjective performance evaluation metric and the optimisation target for the 3D point cloud encoder.
(2) To develop a highly accurate inter-frame block-matching prediction approach for 3D point clouds considering its intrinsic temporal dependency.
(3) To investigate the application of Deep Neural Networks (DNNs) in 3D point cloud compression to reconstruct high-quality 3D point clouds.
Funders
This project is sponsored by the Royal Society and its partner – National Natural Science Foundation of China (NSFC) through the International Exchanges Scheme – Cost Share Programme, which is designed to offer a flexible platform for UK based scientists to interact with the best scientists around the world.
