Motion Vector Prediction Using Non-Adjacent Blocks

The patent segments the prediction block selection into two categories: adjacent blocks and non-adjacent blocks. The motion vector prediction process is divided into checking adjacent blocks first, then selectively incorporating non-adjacent blocks from the buffer when adjacent blocks are insufficient. This segmentation allows the system to maintain simplicity for common cases while improving accuracy when needed.

The patent applies local quality by treating different block positions differently. Adjacent blocks are given priority as they are more likely to have similar motion characteristics. Non-adjacent blocks are only considered when adjacent blocks are unavailable, and only within a restricted buffer range. This local quality approach optimizes prediction accuracy where it matters most while controlling complexity.

The patent performs preliminary action by pre-storing motion information from previously decoded blocks in a buffer before the actual prediction process. This buffer is maintained and updated as decoding progresses, so when non-adjacent blocks are needed for prediction, the required motion information is already available in the buffer, preventing information loss.

The buffer acts as an intermediary between previously decoded blocks and the current prediction process. It stores and manages motion information from non-adjacent blocks, allowing the predictor to access this information without direct dependency on the decoding order. This intermediary buffer prevents information loss by maintaining a restricted set of relevant motion data.