Title: Edge-Oriented Geometric Partitioning for Versatile Video Coding

Abstract: To enhance coding efficiency, the geometric partition (GEO) mode was introduced for the forthcoming VVC standard. This mode offers 140 distinct partition candidates, necessitating the explicit signaling of the optimal GEO mode index. To mitigate the overhead associated with this index and boost coding performance, we introduce a GEO mode prediction strategy. This approach leverages the varying structural traits of different coding units (CUs) alongside the correlations found in spatially adjacent and temporally collocated blocks by constructing a Most Probable Mode (MPM) list. Drawing on the strong correlation observed between partition modes and object boundaries, we propose an edge-directed geometric partition scheme that builds the MPM list using spatio-temporal edge data. When compared to VTM-6.0, our method yields an average objective BD-rate reduction of 0.58% and 1.00% in RA and LDB configurations, respectively. Furthermore, the technique enhances the visual fidelity of object boundaries.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC