Temporal Motion Vector Prediction Using Collocated Blocks

The patent extracts only the necessary motion vector information from previously decoded blocks rather than storing complete motion vector data for all reference pictures. By deriving motion vectors selectively from collocated blocks in previously decoded pictures, the system obtains sufficient prediction accuracy without the overhead of comprehensive motion vector storage, thereby reducing power consumption.

The patent applies local quality by deriving motion vectors only for specific collocated blocks that are relevant to the current prediction task, rather than maintaining motion vector information globally for all reference pictures. This localized approach provides sufficient accuracy for temporal motion vector prediction while minimizing the energy required for storage and retrieval operations.

The patent extracts only the essential motion vector data from previously decoded blocks that is needed for temporal prediction, rather than storing complete motion vector sets for all reference pictures. This selective extraction maintains sufficient derivation accuracy while significantly reducing the memory resources required to store motion vector information.

Instead of storing motion vectors and retrieving them for prediction, the patent inverts the approach by deriving motion vectors on-demand from previously decoded blocks during the decoding process. This inversion eliminates the need for large storage capacity while maintaining the ability to obtain motion vector information when needed for accurate temporal prediction.