Method for coding and an apparatus

What is this patent about?

’105 is related to the field of video compression, specifically the efficient encoding and decoding of motion information. In modern hybrid video codecs, motion vectors are often predicted from previously processed neighboring blocks to reduce the amount of data required to describe movement between frames. A common challenge in this field is the computational overhead and potential for decoder drift when managing a list of motion vector candidates, particularly when redundant or identical candidates are included in the prediction set.

The underlying idea behind ’105 is to streamline the construction of a motion vector merge list by performing a selective, location-based redundancy check rather than an exhaustive comparison of all candidate pairs. By determining a specific subset of candidates to compare based on the geometric position of the current block’s neighbors, the system can effectively prune duplicate motion information with significantly lower complexity. This approach ensures that the encoder and decoder remain synchronized even when certain reference data is unavailable, preventing the propagation of errors through the video sequence.

The claims of ’105 focus on a method and apparatus for generating a merge list by selecting a potential spatial motion vector candidate and comparing its motion information against a specific subset of spatial motion vector prediction candidates. This comparison is triggered by the spatial location of the block associated with the candidate and is designed to avoid a full pairwise comparison of the entire candidate set. The claims specifically cover the use of an equivalence check or similarity metric to identify redundant motion information, ensuring that only unique or necessary candidates are signaled for the prediction unit.

In practice, the invention operates during the inter-prediction phase of video coding, where the system evaluates neighbors such as those located to the left, above, or cornerwise to the current block. Instead of checking every neighbor against every other neighbor, the logic applies specific rules—for instance, only comparing a candidate to its immediate predecessor in a defined order or based on whether the current block is a specific partition of a larger coding unit. This limited number of motion information comparisons reduces the number of operations required to finalize the merge list, which is critical for real-time hardware implementations.

This approach differs from prior solutions that either performed exhaustive redundancy checks, which are computationally expensive, or lacked robust mechanisms to handle missing temporal data. By anchoring the pruning logic to the spatial location of the block, the invention provides a deterministic way to build the merge list that is both efficient and resilient to data loss. This ensures that the final signaled index refers to the same motion candidate at both the encoder and decoder, maintaining high picture quality while minimizing the bit rate overhead associated with motion vector signaling.