Motion Estimation Regions for SbTMVP Merge Candidate Construction

The patent segments the video block into multiple subblocks and constructs merge candidate lists separately for each subblock. This segmentation allows the use of subblock-based temporal motion vector prediction (SbTMVP) modes, which improve encoding efficiency by capturing local motion variations in different parts of the block, while the segmented approach itself manages complexity by breaking down the overall construction process into manageable subblock-level operations.

The patent applies local quality by using different prediction modes (spatial merge candidates vs. temporal SbTMVP candidates) for different subblocks based on their local characteristics. Each subblock can select the most appropriate merge candidate from its local candidate list, allowing the encoding to adapt to local motion patterns and improve overall compression performance without requiring a completely new complex framework.

By segmenting the candidate list construction into subblock-specific lists, the patent achieves higher effective prediction accuracy for each subblock without requiring an excessively long overall candidate list. Each subblock gets tailored candidates from its local spatial and temporal contexts, improving precision while the segmentation naturally limits the complexity of each individual subblock's candidate set.

The patent changes the parameters of merge candidate construction by introducing subblock-level granularity and SbTMVP mode indicators. Instead of using a single uniform candidate list for the entire block, the system varies the candidate selection parameters at the subblock level, allowing more precise motion prediction while managing complexity through parameterized candidate generation rather than brute-force enumeration.

The patent performs preliminary action by pre-determining spatial neighboring blocks and their merge candidates before final merge candidate list construction. The encoder/decoder checks whether spatial neighboring blocks are available and pre-computes their merge candidates, which are then used to generate temporal SbTMVP candidates. This preliminary preparation reduces the complexity of the main SbTMVP determination process and improves compression performance by having ready-to-use candidates.

The system uses self-service by having spatial neighboring blocks serve the current block's merge candidate needs. The spatial neighboring blocks' merge candidates automatically become the temporal SbTMVP candidates for the current block, eliminating the need for separate complex temporal prediction computations. This self-service mechanism improves compression performance while keeping the determination process relatively simple.

The patent applies beforehand cushioning by checking the availability of spatial neighboring blocks and their merge candidates before relying on them for SbTMVP mode. The encoder/decoder determines whether spatial neighboring blocks exist and whether they are within valid Motion Estimation Regions before using them as sources for temporal prediction candidates. This preparatory validation ensures reliability by avoiding references to unavailable or invalid blocks, while still allowing high accuracy when valid neighbors are available.

The patent uses an intermediary approach by introducing a validity check mechanism that mediates between the desire to use spatial neighboring blocks for accurate temporal prediction and the need to ensure their availability. The system checks whether spatial neighboring blocks are valid intermediaries (within MER, properly decoded) before using them to generate temporal SbTMVP candidates, ensuring both accuracy and reliability in the prediction process.