Prediction Image Correction for Accurate Block Boundary Filtering

The patent changes the parameter used for determining filter application from continuity evaluation alone to a combination of prediction accuracy evaluation and continuity evaluation. By introducing prediction accuracy as a new parameter and setting different thresholds (first threshold for prediction accuracy, second threshold for continuity), the system dynamically adjusts whether filter processing is applied, thereby resolving the contradiction between maintaining continuity and preserving prediction image quality.

The patent applies local quality by making the filter processing conditional rather than uniform. Different regions of the inter-predicted image receive different treatments based on local prediction accuracy and continuity characteristics. Specifically, filter processing is applied only to regions where prediction accuracy is below the first threshold and continuity is below the second threshold, while other regions skip filtering, thus improving encoding efficiency without compromising necessary boundary correction.

The patent applies preliminary action by performing prediction accuracy evaluation using multiple reference images before the filter processing decision is made. This preliminary evaluation establishes the prediction accuracy metric that guides subsequent filter application, allowing the system to optimize filter usage based on pre-computed accuracy information rather than making decisions during the filtering stage itself.