Weighted Intra Prediction for Video Coding Accuracy

The patent applies local quality by using different weights for different neighboring pixels based on their spatial relationship to the current pixel. Specifically, pixels closer to the current pixel are assigned higher weights, while distant pixels receive lower weights. This creates a non-uniform weighting scheme where each pixel's contribution to the prediction is optimized for its local position, thereby improving prediction accuracy without requiring a complete redesign of the prediction framework.

The patent changes the parameter of pixel weighting from uniform (conventional averaging) to variable/position-dependent weights. By introducing position-based weight parameters, the prediction method adapts to the spatial characteristics of the image data, allowing closer pixels to have greater influence on the prediction. This parameter change directly addresses the accuracy issue while maintaining computational feasibility through systematic weight assignment.

By applying local quality through position-dependent weighting, the patent reduces the magnitude of difference information between the current block and predicted block. Pixels closer to the current pixel contribute more accurately to the prediction, thereby minimizing the residual error. This leads to better compression efficiency as less information needs to be encoded to represent the differences.

The patent converts the potentially harmful effect of simple averaging (which creates large prediction errors) into a benefit by systematically introducing position-based weights. What was originally a limitation (simple averaging) is transformed into an opportunity to apply intelligent weighting schemes that exploit spatial correlations in the image data, thereby reducing prediction errors and improving compression performance.

The patent implements local quality by calculating weights based on the spatial position of each neighboring pixel relative to the current pixel. This approach focuses computational effort where it is most needed - on pixels that have the greatest influence on the prediction (those closest to the current pixel). The weighting scheme naturally prioritizes computationally important operations while reducing effort on less significant pixels.

By changing from uniform weighting to position-dependent weighting parameters, the patent improves prediction accuracy through a systematic mathematical approach. The weight parameters are derived from pixel coordinates, creating a consistent and repeatable calculation process that, while more complex than simple averaging, remains computationally tractable through established mathematical operations.