Virtual Intra Prediction Mode Extension for Adaptive Inverse Transforms

The patent changes the parameter representation by introducing direction indices derived from virtual intra prediction modes to select transform types. Instead of using traditional fixed transform selection, the system dynamically selects transforms based on directional parameters (first and second directions) obtained from the prediction block, allowing adaptive transform choice that improves compression efficiency while managing complexity through parameter-based selection.

The patent performs preliminary action by pre-defining mapping relationships between direction indices and transform indices. The system prepares candidate inverse transform index lists and establishes mapping tables in advance, so that during actual video decoding, the transform selection can be quickly determined through index mapping rather than complex real-time analysis, thus reducing operational complexity while maintaining high compression efficiency.

The patent segments the transform selection process by dividing it into distinct stages: obtaining the prediction block, determining first and second directions separately, selecting transform types based on each direction, and finally performing the inverse transform. This segmentation allows the system to handle multiple directions systematically, improving video quality through multi-directional analysis while managing processing time by breaking down the complex operation into manageable sequential steps.

The patent applies partial action by selectively applying different inverse transform types to different regions or aspects of the video block based on directional characteristics. Instead of uniformly applying complex multi-directional transforms to the entire block, the system uses the first and second directions to select appropriate transforms for specific portions or frequency components, achieving enhanced video quality where needed while reducing overall processing time through targeted rather than exhaustive transformation.

The patent implements self-service by having the system automatically calculate histograms of directional characteristics from the prediction block itself and use these histograms to determine the mapping between direction indices and transform indices. The video content provides its own statistical information through the histogram calculation, eliminating the need for external parameter input or manual configuration, thus achieving high content adaptability while keeping the process self-contained and efficient.

The patent uses feedback mechanisms by calculating histograms based on the actual directional information present in each video block and using this feedback to adaptively select the appropriate transform indices. The system continuously monitors the directional characteristics through histogram computation and adjusts the transform selection accordingly, creating a closed-loop system that adapts to different video content types (e.g., horizontal, vertical, or diagonal patterns) while maintaining computational efficiency through iterative refinement rather than exhaustive search.