Inter Prediction Reference Unit Selection for Video Compression

The patent segments the reference unit determination process into distinct stages: identifying candidate neighboring units, selecting candidate units based on criteria (boundary length, size, depth value), and finally determining the reference unit from the candidate set. This segmentation reduces complexity by breaking down the overall task into manageable sub-tasks with clear decision rules for each stage.

The patent performs preliminary actions by pre-identifying and selecting candidate units from neighboring units before the actual reference unit determination. Units are pre-filtered based on boundary length, size, and depth value criteria, creating a reduced candidate set that simplifies the subsequent reference unit selection process and improves encoding efficiency.

The patent applies local quality by evaluating different neighboring units based on their specific local characteristics: boundary length adjacent to the current unit, size of the unit, and depth value in the hierarchy. Each unit is assessed according to its local properties rather than applying a uniform criterion, allowing accurate selection while maintaining reasonable processing time through targeted evaluation.

The patent implements partial action by not evaluating all possible neighboring units exhaustively. Instead, it selectively evaluates units based on predefined criteria (boundary length threshold, size threshold, depth value threshold) to identify a sufficient candidate set. This partial evaluation approach achieves adequate prediction accuracy without the excessive processing time that would result from exhaustive search of all possible units.