Intra Template Matching Vector Refinement for Video Block Prediction

The patent applies preliminary action by generating merge candidate lists and performing reference block vector refinement in advance during the encoding process. The encoder pre-computes multiple candidate reference block vectors and selects the optimal one before final encoding, allowing the decoder to simply follow the same procedure without real-time complex computations. This advance preparation improves coding accuracy while managing device complexity through systematic pre-processing.

The patent implements dynamics by introducing adaptive reference block vector refinement that dynamically adjusts the prediction process. The system evaluates multiple merge candidates and selectively refines reference block vectors based on specific conditions and metrics, rather than applying a static uniform process. This dynamic approach improves coding accuracy by adapting to different block characteristics while controlling computational overhead through selective refinement.

The patent applies segmentation by dividing the reference block vector refinement process into distinct stages: generating multiple merge candidates, evaluating each candidate, selecting optimal candidates, and performing refinement. This segmented approach allows the system to process different blocks using different levels of refinement based on their characteristics, improving overall compression performance while reducing average processing time by avoiding full refinement for all blocks.

The patent implements partial action by performing reference block vector refinement selectively rather than universally. The system applies full refinement only to blocks where it provides significant benefit, while using simpler methods for other blocks. This partial approach improves compression performance for critical blocks while keeping overall processing time manageable by avoiding excessive refinement everywhere.