LIghtweight Texture-Guided Fast Partition Method for Luma and Chroma Intra Coding in VVC

Despite the notable advancements in encoding efficiency by VVC, its increased complexity hinders real-time and practical applications. We propose a lightweight online learning-based CU partition acceleration method, which identifies redundant partition types by texture features. The model accommodates CUs of varying sizes and considers both luma and chroma components, providing encoding acceleration with strong resistance to overfitting. Experimental results on VTM-IO.0 validates the efficacy of our approach, which achieves 44.7$ time savings while maintaining a slight BDBR loss of0.96$ in All Intra configuration, and preserving almost lossless subjective quality.