Affine Merge Motion Vector Blending to Reduce Coding Complexity

The patent divides the current block into multiple sub-blocks and performs motion vector derivation separately for each sub-block using regression models. This segmentation allows the system to apply complex regression-based methods only where needed (in regions with significant motion variation) rather than uniformly across the entire block, thereby improving coding efficiency for complex regions while limiting computational overhead to only necessary areas.

The patent applies regression-based motion vector derivation selectively rather than universally. It uses regression models only for sub-blocks where motion complexity warrants the additional computation, while simpler regions may use traditional motion compensation methods. This partial application of the complex algorithm maintains coding efficiency for difficult regions while reducing overall computational complexity.

By segmenting the block into sub-blocks and applying regression-based derivation only to those requiring high motion accuracy, the system achieves precise motion estimation for complex regions without spending excessive processing time on uniform regions that can be handled by simpler methods.

The patent performs preliminary classification or assessment of sub-block motion characteristics before applying the computationally intensive regression-based derivation. This preliminary action identifies which sub-blocks require high-precision motion vector derivation, allowing the system to prepare and apply regression models only where necessary, thus improving motion accuracy for critical regions while minimizing overall processing time.