CU-Level Bi-Prediction Weighting for Lower-Complexity Video Coding

The patent applies parameter changes by providing a list of predefined candidate weights and allowing the encoder/decoder to select from these discrete weight values rather than using arbitrary continuous weights. This discretization of the weight parameter space reduces the complexity of weight selection while maintaining compression efficiency. The sign value parameter is also used to indicate the direction of larger weight values, further simplifying the representation.

The patent implements preliminary action by pre-defining a list of candidate weights before the actual prediction process. These candidate weights are prepared in advance and stored in a lookup table or similar structure, allowing the encoder/decoder to simply retrieve and apply the appropriate weight without performing complex weight calculation during the prediction operation. This pre-computation approach significantly reduces real-time processing complexity.

The patent applies copying by using predefined candidate weights that are copied from a lookup table or pre-computed structure rather than calculating weights in real-time. The encoder/decoder simply retrieves the appropriate candidate weight from the pre-defined list, which copies the optimal weight values without requiring complex computational operations during the prediction process. This copying approach maintains compression performance while dramatically reducing computational complexity.

The patent uses cheap short-living objects by employing simple sign values (e.g., +1 or -1) to indicate the direction of larger weight values instead of using complex weight calculations. These sign values are inexpensive to compute and store, and they provide sufficient information to reconstruct the appropriate weight application without requiring expensive computational operations. The complexity is traded off in favor of simplicity and speed.

The patent applies preliminary action by pre-computing and storing the relationship between pixel differences and optimal weights in a lookup table or pre-defined structure. During the actual prediction process, the system simply retrieves the appropriate weight based on the pixel difference magnitude from this pre-computed structure, rather than performing complex weight derivation calculations in real-time. This pre-computation approach maintains prediction accuracy while significantly reducing processing time.

The patent uses cheap short-living objects by replacing complex pixel-level weight derivation with simple sign value indicators that are inexpensive to compute. Instead of performing full pixel-level optimization, the system uses these simple sign values to indicate the direction of weight adjustment, which are much faster to compute and store while providing sufficient prediction accuracy. The time-consuming pixel difference analysis is done in a simplified manner that balances accuracy and speed.