[Paper] Hybrid Spatial and Deep Learning-based Point Cloud Compression with Layered Representation on 3D Shape

Abstract

It is expected that the shapes of real-world objects such as buildings and people can be sensed, stored as point clouds, and utilized. For efficiently storing and transmitting a huge amount of point cloud data, point cloud compression methods based on deep learning have been studied. In order to grasp an overview or details of a desired building or person on a display, it is an important function to extract whole or a desired part of the point cloud from the compressed data and represent the characteristic shape of the object. In this paper, a hybrid point cloud encoding method is proposed, which consists of a layered structuring that presents the main features of the point cloud with various number of points and an efficient block-wise encoding by combining deep learning.