Motion Vector Candidate Selection for Efficient Video Decoding

The patent applies dynamics by making the prediction method adaptive rather than static. The encoder dynamically selects between two different prediction approaches (conventional median-based prediction and new candidate-based prediction) based on the actual motion characteristics of the video content. This allows the system to optimize prediction accuracy for each specific block while maintaining efficiency through conditional selection rather than always using the more complex method

The patent changes the prediction parameter by introducing multiple candidate motion vectors (left, above, above-right candidates) and using deviation evaluation metrics to select the best prediction method. Instead of always using the simple median of three candidates, the system evaluates the distribution characteristics of these candidates and switches between prediction strategies based on measured deviation values, thereby adapting to different motion patterns

The patent segments the prediction process into distinct evaluation stages: calculating deviation metrics for each candidate block, comparing these deviations against thresholds, and selectively applying different prediction methods based on the evaluation results. This segmented approach with clear decision boundaries (thresholds) manages complexity by breaking down the adaptive selection process into manageable, conditional steps rather than a monolithic complex algorithm

The patent implements feedback through deviation evaluation, where the encoder measures the actual distribution characteristics of candidate motion vectors and uses this feedback to determine the optimal prediction method. The deviation metrics provide quantitative feedback about the suitability of different prediction approaches, enabling the system to adaptively optimize compression efficiency while maintaining controlled complexity through objective decision criteria