Lossy Compression of Point Clouds Using Coding Rate Reduction Compressed Sparse Representation

Point cloud compression is a technique that aims to address the challenge of storing and transmitting large-scale 3D data by reducing the size of point cloud data while maintaining sufficient useful information. However, the large amount of data in point clouds and their irregular spatial arrangement complicates storage and transmission. To address this issue, a novel lossy point cloud compression method is proposed in this paper. Our strategy treats point cloud compression as an iterative optimization process, where each layer performs an optimization step to achieve the sparse compression goal. An autoencoder structure is used to enhance the point cloud compression task with coding rate reduction transformer and geometric attribute information. In the encoding phase, a novel self-encoder is used to learn the compressed and sparse representation and geometric features of the point cloud to be used as a compressed embedding of the point cloud. In the decoding stage, point-wise split deconvolution is designed for upsampling the point cloud in order to better reconstruct the point cloud. Experimental results show that the proposed method achieves better reconstruction quality and geometry at the same bit rate as other state-of-the-art compression methods.