Exponential Partitioning for Video Block Inter Prediction

The patent uses exponential curves instead of straight lines to partition blocks, allowing the partition boundary to adapt to object boundaries and motion patterns. The curve parameters a and b are derived from block dimensions and position, enabling straightforward implementation while achieving more accurate motion compensation predictions by reducing the mismatch between partition boundaries and actual object boundaries.

The patent changes the partitioning approach from fixed straight lines to parameter-based exponential curves. The parameters a and b are calculated based on block width W and height H, allowing the curve to adapt to different block sizes and positions. This parameterization maintains implementation simplicity while significantly improving motion compensation prediction accuracy by enabling flexible boundary alignment.

The patent simplifies the exponential curve partitioning process by parameterizing the curve with two values (a and b) that can be directly computed from block dimensions and position. The parameter a is calculated as (2^k - 1) / (2^k + 1) where k is derived from block size, and parameter b is computed from the block position and dimensions. This parameterization approach maintains low implementation complexity while achieving superior compression efficiency through accurate object boundary following.

The patent employs exponential curves to partition video blocks, allowing the boundary to follow object contours more accurately than straight lines. The curve is defined by parameters that are efficiently computed from block metadata, avoiding the need for complex geometric calculations. This approach improves compression efficiency by reducing motion compensation errors while keeping the partitioning process computationally lightweight through closed-form parameter calculations.