Planar Intra Prediction Using Adaptive Reference Samples
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video coding standards, such as HEVC, face challenges in achieving efficient compression of video data, particularly in capturing complex edge directions and textures, leading to suboptimal prediction modes and increased computational complexity.
Innovation Solution
The VVC standard extends angular intra prediction modes to 65 directions, introduces planar and direct chroma prediction modes, employs a most probable mode list (MPM) for mode selection, and utilizes decoder-side intra mode derivation methods like template-based intra mode derivation (TIMD) and position-dependent intra prediction combination (PDPC) to refine predictions, along with advanced transform techniques like Multiple Transform Selection (MTS) and Low-Frequency Non-Separable Transform (LFNST).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing video coding standards (HEVC) are used, then device complexity is reduced, but compression efficiency and prediction accuracy deteriorate
Solution Approach 1:
The patent segments the prediction mode selection process into multiple stages: first selecting from a most probable mode list (MPM) containing angular modes, then refining selection based on planar mode characteristics. This segmentation allows the system to achieve high compression efficiency through sophisticated prediction while managing complexity by structuring the decision process in manageable steps rather than requiring a single complex algorithm
Solution Approach 2:
The patent introduces planar mode parameters as a new dimension to the prediction mode selection process. By changing the parameter space to include both angular modes and planar modes with specific parameters (such as reference sample indices and prediction directions), the system achieves improved compression efficiency while the parameterization provides a systematic way to manage the increased complexity
2Measurement precision
If angular intra prediction modes are extended to 65 directions, then prediction accuracy for complex edge directions is improved, but device complexity increases
Solution Approach 1:
The patent segments the 65 angular prediction modes into a structured most probable mode list (MPM) that is derived from neighboring block modes. Rather than implementing all 65 modes uniformly, the system segments the mode selection based on local image characteristics and neighboring context, achieving high edge direction capture accuracy where needed while reducing complexity by not uniformly applying all modes everywhere
Solution Approach 2:
The patent makes the prediction mode list dynamic by deriving it from neighboring block modes rather than using a fixed set. The MPM adapts to the local image content and edge directions present in neighboring blocks, allowing the system to capture complex edge directions accurately when present while avoiding the complexity of processing all 65 modes in all cases
3Productivity
If decoder-side intra mode derivation methods (TIMD, PDPC) are used, then compression efficiency is improved, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by performing template-based intra mode derivation (TIMD) and position-dependent intra prediction combination (PDPC) at the decoder side using reference data that has already been processed. The template derivation uses previously decoded neighboring blocks to predict the current block mode, and PDPC combines predictions from multiple positions. This preliminary processing of reference data enables improved compression efficiency while the computational complexity is managed by operating on already-decoded data rather than requiring complex real-time analysis
4Manufacturing precision
If multiple transform selection (MTS) and low-frequency non-separable transform (LFNST) are applied, then texture handling is improved, but device complexity increases
Solution Approach 1:
The patent applies dynamics by making the transform selection adaptive rather than fixed. The Multiple Transform Selection (MTS) allows dynamic choice between different transform types (e.g., DCT, DST, ADST) based on the characteristics of the current block and neighboring blocks. The Low-Frequency Non-Separable Transform (LFNST) dynamically adjusts the transform kernel selection based on frequency content. This dynamic adaptation improves texture handling quality by selecting the most appropriate transform for each case while managing complexity through data-driven selection rather than requiring all transforms to be applied uniformly
Data Source
AI summary
An input video or video stream may be obtained or received. The input video or video stream may include a plurality of video frames, and each frame may be divided into a plurality of blocks. A current block of the plurality of blocks may be predicted using a planar mode. Depending on which planar mode is used, different reference samples may be used for predicting a current sample in the current block.