Image Prediction Search for High-Efficiency Video Encoding

The patent segments the prediction process into distinct modes (inter-prediction, intra block copy, intra template matching) and further divides the motion vector search into multiple stages with different search patterns. This segmentation allows the system to select appropriate complexity levels for different image regions and prediction scenarios, improving prediction accuracy without uniformly increasing computational complexity across all blocks.

The patent implements partial action by applying exhaustive motion vector searches only when necessary (e.g., when template matching indicates high similarity potential) while using simplified search methods for other cases. This selective approach ensures high prediction accuracy for difficult-to-compress regions while maintaining lower computational complexity for easier regions, effectively resolving the contradiction between accuracy and complexity.

The patent employs periodic action through its multi-stage motion vector search methodology, where different search patterns are applied in sequence (e.g., coarse search followed by fine search). This periodic approach allows the encoder to periodically refine prediction precision at strategically chosen stages rather than continuously performing exhaustive searches, thereby reducing overall encoding time while maintaining high prediction precision through cumulative refinement.

The patent applies preliminary action by performing template matching and preliminary motion vector estimation before the final bidirectional prediction and motion vector search. This preliminary action pre-identifies promising search regions and candidates, allowing the subsequent bidirectional prediction to focus computational resources on refined searches that yield high precision results more quickly, thus reducing total encoding time.