Color Component Prediction Using Reference Pixel Subset Selection

The patent extracts only the necessary reference pixels from the full reference pixel set to construct the prediction model. By identifying and removing unnecessary pixels through classification (e.g., using gradient information or position-based selection), the method retains only those pixels that contribute meaningfully to prediction accuracy, thereby reducing calculation complexity while maintaining model reliability.

The patent applies local quality by treating different regions of the reference pixel set differently. Instead of uniformly using all pixels, the method selectively weights or includes pixels based on their local characteristics (such as gradient magnitude, position relative to the current block, or variance). This allows the prediction model to focus on locally relevant information, reducing overall computational load while preserving accuracy where it matters most.

The patent extracts a reduced subset of reference pixels from the complete reference pixel set. By using selection criteria (such as gradient-based filtering or position-based sampling), the method retrieves only the essential pixels needed for accurate prediction, thereby significantly reducing the memory bandwidth required to load and process reference data while maintaining model reliability.

The patent applies partial action by using only the necessary portion of the reference pixel set rather than the entire set. The method determines that a subset of pixels (partial action) is sufficient to achieve the required prediction accuracy, avoiding the excessive use of memory bandwidth that would result from processing all available pixels.

The patent extracts and removes exceptional or outlier samples from the reference pixel set before constructing the prediction model. By using classification criteria (such as gradient thresholding, position-based filtering, or statistical outlier detection), the method identifies and excludes pixels that would negatively impact prediction accuracy, thereby improving measurement precision while maintaining model robustness through the use of high-quality remaining samples.

The patent applies local quality by evaluating each reference pixel's individual characteristics and treating them differently based on their quality. High-quality pixels (those that conform to expected patterns and provide reliable prediction information) are included in the model, while low-quality or exceptional samples are excluded or down-weighted. This selective approach improves overall prediction accuracy by ensuring that only reliable local information contributes to the model.