Mapping-Based Video Coding for Diverse Sample Distributions

The patent applies parameter changes by transforming sample values through mapping functions that adapt to diverse distributions. The encoder determines mapping parameters (such as scaling and offset factors) based on the statistical characteristics of the current block's sample values, and applies these parameters to transform the sample values into a standardized range. This allows the prediction process to operate on uniformly distributed values regardless of the original distribution, thereby reducing prediction error and improving encoding performance.

The patent reduces data amount by applying mapping transformations that normalize sample value distributions before encoding. By transforming diverse distributions into a standardized range using learned mapping parameters, the encoder can achieve better prediction accuracy with fewer bits. This allows high-quality encoding of large-resolution, high-frame-rate video without proportionally increasing the data amount, as the mapping process improves compression efficiency.

The patent resolves the distribution mismatch between reference samples and encoded samples by applying mapping transformations to the encoded samples. The encoder determines mapping parameters based on the current block's sample value distribution and applies these parameters to transform the decoded samples into a distribution that matches the reference samples. This allows the predictor to effectively utilize reference images with different original distributions, maintaining prediction performance while preserving adaptability to various reference image sources.