Control-Point Motion Refinement from Coded Samples for Video Coding

The patent divides the video block into multiple subblocks, each with its own motion compensation parameters. This segmentation allows different regions to handle complex motions independently, improving both coding efficiency and effectiveness for zoom, rotation, and perspective transformations.

The patent introduces dynamic motion compensation where motion parameters are not fixed but can be refined based on actual content characteristics. The system dynamically adjusts motion models and parameters to adapt to complex motion patterns, enhancing coding effectiveness while maintaining efficiency.

By segmenting the video block into subblocks, the patent simplifies the overall motion modeling complexity. Each subblock can use simpler motion models while collectively capturing complex motion patterns, reducing the computational burden compared to a single complex model for the entire block.

The patent applies different motion compensation strategies to different subblocks based on their specific motion characteristics. This local quality approach allows simple regions to use simple models while complex regions receive enhanced modeling, optimizing the balance between prediction accuracy and modeling complexity.

The patent applies refinement processes selectively to only those subblocks or motion parameters that require it, rather than uniformly refining the entire video block. This partial action approach maintains coding effectiveness for complex motion regions while minimizing processing time for simpler regions.

The refinement process dynamically adjusts its intensity and scope based on the complexity of motion detected in different regions. Simple regions skip refinement or use minimal processing, while complex regions receive intensive refinement, optimizing the trade-off between coding effectiveness and processing time.