Interleaved Intra Gradient Filter for Video Coding

The patent segments the 33 directional Intra prediction modes into multiple groups (e.g., first group with modes 2-10, second group with modes 11-19, etc.) and applies the Intra gradient filter selectively to specific groups rather than all modes. This segmentation allows the encoder to improve coding efficiency for modes that benefit most from filtering while avoiding the complexity overhead for modes where filtering provides minimal gain, thus resolving the contradiction between coding efficiency and device complexity.

The patent applies different filtering strategies to different local regions of the prediction mode space. Specifically, it applies Intra gradient filtering to certain directional modes (e.g., modes 2-10 and 11-19) while using alternative methods (e.g., bi-directional prediction or no filtering) for other modes (e.g., modes 20-33). This local quality approach ensures that filtering complexity is concentrated only where it provides the most benefit, balancing coding efficiency improvement with complexity reduction.

The patent segments the Intra prediction modes into groups where bi-directional prediction is applied to some groups (e.g., modes 2-19) while other groups (e.g., modes 20-33) use different prediction methods. This segmentation avoids the boundary sample availability issues that arise when bi-directional prediction is applied universally, as certain modes have inherent availability problems that would require complex handling mechanisms.

The patent applies bi-directional prediction partially, only to specific mode groups where it provides significant accuracy improvement, rather than excessively applying it to all modes. This partial action approach achieves the necessary prediction accuracy for critical modes while avoiding the complexity overhead of handling boundary sample availability issues for all modes, thus resolving the contradiction between prediction accuracy and complexity.

The patent segments modes into categories based on boundary sample availability characteristics. It applies Intra gradient filtering only to modes where boundary samples are naturally available or easily obtainable, while using alternative approaches for modes with unavailable boundary samples. This segmentation eliminates the need for complex sample padding mechanisms while maintaining prediction quality for the modes that require filtering.

The patent extracts and identifies the specific subset of modes that benefit from Intra gradient filtering without requiring complex sample padding operations. By taking out only the modes that can be effectively filtered with available boundary samples (e.g., modes 2-19), it avoids the complexity of implementing padding mechanisms while still achieving prediction quality improvement for the appropriate mode set.