Color Attribute Compression for Block Based Representation of Point Cloud
Hideaki Kimata · 2023
Compression of point cloud obtained by sensing real-world objects with LiDAR or RGBD sensors has been studied. Block-based geometry compression methods using deep learning have been presented, however, less studies have been reported on compression of attribute information such as colors. In this paper, an efficient encoding of color attribute information is proposed for block-based geometry compression, which has an advantage that parts of point cloud are processed in parallel. The proposed method encodes color information as an image projected onto a surface, block by block, in order to achieve better subjective quality of the rendered image. A deep learning-based image compression method for the projected image is also studied. The overall efficiency is discussed in this paper.