Edge Optimization Algorithm for Geometry Maps in AV1-Enabled V-PCC

Point cloud is a three-dimensional data representation based on points and is widely applied in the field of 3D vision technology to describe the geometry of objects or scenes. They serve as a crucial data storage format. However, due to the large volume of point cloud data, compression techniques are essential for reducing its size. To address this, the Moving Picture Experts Group (MPEG) developed the Video-based Point Cloud Compression (V-PCC) process specifically for dynamic point cloud data, utilizing High-Efficiency Video Coding (HEVC) as the encoder. Despite these advancements, point cloud compression often results in the loss of edge points, which leads to the formation of holes, the appearance of compression artifacts, and a subsequent degradation in visual quality. This study employs the AOMedia Video 1 (AV1) encoder for point cloud compression and introduces an effective method to mitigate compression artifacts. By leveraging the characteristics of occupancy maps generated during the V-PCC process, this study identify the edges between patches in 2D projected point cloud images from different projection angles by a distance transform algorithm. Along these edges, distortion values in the geometry maps before and after compression are recorded, and pixels with significant distortion are compensated. The optimized geometry map is then re-encoded by AV1 and further processed through the V-PCC decoding workflow, ultimately producing a higher-quality reconstructed point cloud. Experimental results demonstrate that the proposed method achieves an average improvement of 2.63% and 2.12% in point-to-point and point-to-plane MSE, respectively. Furthermore, in terms of BD-rate results, the compression efficiency for point-to-point and point-to-plane PSNR improves by an average of -3.0% and -2.2%, respectively, thereby further enhancing the subjective visual quality of the reconstructed point cloud.