Content-Aware Chroma Prediction in Video Bitstream Conversion

The patent implements dynamic selection of chroma prediction tools (CCLM, PDPC, DBV) based on content detection and mode flags. The encoder and decoder adaptively enable or disable these tools for different video units based on detected content characteristics, allowing the system to optimize coding efficiency dynamically rather than using a fixed approach. This resolves the contradiction by making the coding system flexible and adaptive to different video content types.

The patent introduces multiple prediction parameters including CCLM mode flags, PDPC mode flags, DBV mode flags, and various prediction mode indices (intra_chroma_pred_mode). These parameters are dynamically adjusted and selected based on content detection results, enabling the system to change coding parameters adaptively to improve both coding efficiency and performance across different video content scenarios.

The patent applies different chroma prediction tools and modes to different local regions (video units) based on content detection. Instead of uniformly applying complex prediction tools across the entire video stream, the system selectively enables CCLM, PDPC, or DBV modes only for specific video units where they are most beneficial, as determined by content characteristics and mode flags. This localizes the complexity to where it provides maximum performance benefit.

The patent implements a hierarchical approach where basic chroma prediction is always applied, and additional prediction tools (CCLM, PDPC, DBV) are selectively added only when content detection indicates they would provide benefit. This partial application of advanced tools avoids the excessive complexity of always using all available tools while still achieving performance improvements where needed.

The patent performs content detection and determines the appropriate chroma prediction mode in advance during the encoding process, before actual chroma prediction is performed. The content detection analyzes video unit characteristics and pre-determines which prediction tools (CCLM, PDPC, DBV) should be enabled, storing this information in mode flags. This preliminary action allows the actual prediction to proceed efficiently without repeated detection overhead during decoding.

The patent implements a feedback mechanism where content detection results from encoding are used to set mode flags that guide both encoding and decoding operations. The decoder uses these pre-determined mode flags to quickly identify which prediction tools were applied without needing to perform its own content detection, thus maintaining coding efficiency improvements while minimizing the processing time penalty to only the encoding stage.