Occupancy‐vector‐based eligibility method for RAHT inter‐prediction of point cloud compression

Abstract Region‐adaptive Hierarchical Transform inter‐frame prediction in Geometry‐based Point Cloud Compression (G‐PCC) effectively reduces redundancy without requiring motion vectors. However, conventional approaches rely only on direct current coefficient thresholds or spatial matching, which are considered insufficient for accurate evaluation of attribute prediction. To improve this prediction accuracy, we propose an occupancy‐vector‐based eligibility method that characterizes the geometric structures of both reference and target transform nodes using occupancy vectors and computes the cosine similarity outcomes between these transform nodes. The proposed method determines whether inter‐prediction should be applied or not, which ensures that it is performed only when local geometric structures are sufficiently similar. Our approach minimizes unnecessary predictions for dense dynamic point clouds and thus improves the reliability of inter‐prediction within the transform framework, consequently enhancing the compression efficiency of G‐PCC. Experimental results show up to 3.4% average Bjøntegaard Delta rate savings in lossy compression, with marginal impact in lossless compression.