Method for coding and an apparatus

What is this patent about?

’714 is related to the field of video compression, specifically the mechanisms used to predict motion vectors during the encoding and decoding of video sequences. In modern hybrid video codecs, motion information for a current block is often predicted from previously coded neighboring blocks to reduce the amount of data that must be transmitted. This process involves constructing a list of potential motion vector candidates, known as a merge list, from which an optimal predictor is selected and signaled.

The underlying idea behind ’714 is to streamline the construction of the motion vector candidate list by performing a limited, intelligent redundancy check rather than an exhaustive comparison of all possible candidate pairs. The inventor recognized that comparing every available spatial and temporal candidate to remove duplicates is computationally expensive and can lead to decoding errors if reference data is lost. By using the spatial location of a candidate block to determine a specific, restricted subset of other candidates for comparison, the system can effectively prune the list with significantly lower overhead.

The claims of ’714 focus on a method and apparatus for generating a motion vector prediction list by selecting a potential spatial candidate and determining a specific subset of existing candidates for comparison based on that candidate's location. The independent claims specify that the motion information of the first candidate is compared only with this determined subset, explicitly avoiding a full all-pairs comparison of the entire candidate set. Based on this targeted comparison, the system decides whether to include or exclude the candidate from the final list used for signaling or decoding.

In practice, the invention operates by evaluating neighboring blocks—such as those located to the left, above, or at the corners of the current prediction unit—in a specific sequence. For each neighbor, the codec identifies which previous neighbors are likely to hold identical motion data based on the geometric relationship and the partitioning mode of the coding unit. For example, if a block is split into two vertical units, the second unit may automatically exclude certain neighbors to prevent redundant signaling, ensuring the merge list remains compact and computationally efficient to process.

This approach differs from prior solutions that either performed exhaustive redundancy checks or used fixed candidate sets that did not adapt to the local geometry of the prediction units. By implementing location-based pruning, the invention ensures that the encoder and decoder remain synchronized even in high-complexity scenarios or when temporal data is unavailable. This targeted logic reduces the number of required comparison operations while maintaining high compression efficiency, effectively cutting through the processing bottleneck of candidate list management.