Geometric Intra Prediction Partitioning for Directional Video Blocks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video encoding and decoding technologies struggle to efficiently handle blocks in video frames with varying directional structures, as they often require a single prediction mode that inadequately describes regions with different directionalities, leading to inefficiencies in compression and reconstruction.
Innovation Solution
The proposed method involves splitting a block into at least two partitions using a straight line, performing intra prediction with different modes on each partition, and adjusting sample values along the split line with adaptive weights through a blending process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single intra prediction mode is used for the entire block, then the encoding complexity is low, but the prediction accuracy deteriorates for blocks with varying directional structures
Solution Approach 1:
The current block is divided into multiple sub-blocks, and different intra prediction modes are applied to each sub-block independently. This segmentation allows the encoder to capture varying directional structures within different regions of the block, significantly improving prediction accuracy for complex blocks while maintaining manageable encoding complexity through localized processing.
Solution Approach 2:
Different prediction modes are selected and applied to different sub-blocks based on their local characteristics. Each sub-block can use the most appropriate prediction mode for its specific directional structure, ensuring optimal local prediction quality while the overall block benefits from the combination of locally optimized predictions.
2Measurement precision
If the block is divided into multiple partitions with different prediction modes, then the prediction accuracy improves, but the computational complexity increases
Solution Approach 1:
The block is divided into multiple partitions that can be processed independently with different prediction modes. This segmentation enables parallel processing of sub-blocks, reducing the overall computational burden while maintaining high prediction accuracy through mode diversity.
Solution Approach 2:
The method applies geometric partitioning selectively based on block characteristics. Not all blocks require full geometric partitioning with multiple prediction modes - the technique can be applied partially or conditionally, reducing computational complexity for blocks that don't benefit as much while maintaining high accuracy for blocks that do require it.
3Productivity
If geometric partitioning is applied to all blocks, then the compression efficiency improves, but the encoding overhead increases
Solution Approach 1:
Geometric partitioning is applied selectively rather than universally. The encoder determines whether geometric partitioning is beneficial for each block based on its characteristics, applying the technique only where it provides significant compression benefits. This partial application reduces encoding overhead while maintaining high compression efficiency for suitable blocks.
Solution Approach 2:
The method introduces geometric partition parameters (such as partition orientation and sub-block division) that can be adjusted based on block characteristics. By dynamically changing these parameters rather than applying a fixed partitioning scheme, the encoder achieves better compression efficiency while managing overhead through adaptive parameter selection.
Data Source
AI summary
In one implementation, when coding a block in the intra prediction mode, the block can be split into two sub-partitions by a geometrically located straight line. Each geometric partition within the CU is intra predicted using its own intra mode with its available reference sample. One sub-partition copies and uses the intra prediction mode from the parent block, and another sub-partition uses another implicit or explicit signaled intra prediction mode. After predicting geometric partition, the sample values along the split boundary are adjusted using a blending process with adaptive weights. The geometric partition based intra prediction could be applied for one angular intra prediction mode, or only for one negative-directional intra prediction mode, or only for one specific intra prediction mode (e.g., mode 34). The transform selection or other intra coding tools (i.e., intra sub-partition) can be adapted for the geometric partition based intra prediction.