Video Signal Intra Prediction Using Non-Adjacent Reference Samples

The patent segments the reference samples into multiple groups based on their spatial relationships. Specifically, it divides reference samples into first reference samples (from one direction) and second reference samples (from another direction), allowing the encoder to select appropriate groups for different prediction modes. This segmentation enables more flexible and accurate prediction while maintaining manageable complexity through structured organization.

The patent extends the traditional one-dimensional adjacent reference sample approach by incorporating reference samples from multiple spatial dimensions and directions. It uses both horizontal and vertical reference samples, and introduces the concept of non-adjacent reference samples that are not immediately next to the current block but provide additional spatial context. This multi-dimensional approach significantly improves prediction accuracy for high-resolution content.

The patent implements dynamic selection of reference sample groups based on the current block's characteristics and prediction mode. The encoder adaptively chooses which reference sample groups to use (first reference samples, second reference samples, or both) depending on the specific encoding context. This dynamic approach allows the system to achieve high precision when needed while maintaining faster processing when simpler prediction suffices, thus balancing quality and speed.

The patent changes the parameters of the prediction process by introducing multiple reference sample groups with different spatial characteristics. It varies the composition and weighting of reference samples based on the prediction mode and block type, allowing flexible adjustment of prediction accuracy versus computational cost. This parameter variation enables optimized encoding performance across different video content types.