Unequal-Offset Reference Lines for VVC Intra Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing Multiple Reference Line (MRL) intra prediction in VVC and ECM is inefficient, particularly for positive angular prediction directions, due to the constraint of equal offsets for reference lines, which limits the effectiveness of position-dependent prediction combination (PDPC) operation.
Innovation Solution
Implementing unequal offset reference lines in MRL and template-based MRL (TMRL) intra prediction, allowing for more combinations of reference samples by signaling the best pair of top and left reference arrays using a suitable encoding scheme.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If equal offset reference lines are used in MRL intra prediction, then the reference line structure is simple and easy to implement, but the prediction accuracy is limited particularly for positive angular prediction directions
Solution Approach 1:
The patent applies asymmetry by allowing the top reference array and left reference array to have different offset distances from the target block. Specifically, the top reference array can be at offset M while the left reference array is at offset N, where M ≠ N. This asymmetric configuration enables better adaptation to positive angular prediction directions by providing reference samples at optimally positioned locations, thereby improving prediction accuracy without requiring a completely new reference line structure
Solution Approach 2:
The patent implements local quality by selectively applying different offset configurations to different parts of the reference structure. The top and left reference arrays can have different offsets depending on the prediction direction and block characteristics. This allows the reference line structure to be locally optimized for specific prediction scenarios (particularly positive angular modes) while maintaining the general MRL framework, thus improving prediction accuracy without globally increasing complexity
2Measurement precision
If unequal offset reference lines are implemented to improve prediction accuracy, then more combinations of reference samples become available, but the encoding complexity and computational load increase
Solution Approach 1:
The patent applies parameter changes by modifying the offset parameters of the reference arrays. Instead of fixed equal offsets, the system allows the top reference array offset (M) and left reference array offset (N) to be independently configurable parameters. This enables the encoder to select optimal offset values based on prediction direction and block characteristics, improving prediction accuracy while managing complexity through parameter optimization rather than structural overhaul
Solution Approach 2:
The patent implements dynamics by making the reference array offsets adaptive rather than static. The offset values M and N can be dynamically selected based on the prediction mode, block size, and other coding parameters. This dynamic configuration allows the system to achieve better prediction accuracy for positive angular modes when needed, while falling back to simpler configurations for other cases, thus balancing accuracy improvement with encoding complexity
3Measurement precision
If unequal offsets are used for top and left reference arrays, then rate-distortion performance is improved, but the number of possible reference sample combinations increases significantly
Solution Approach 1:
The patent applies parameter changes by optimizing specific offset parameters (M and N) to achieve the best rate-distortion performance. Rather than allowing all possible offset combinations, the system identifies and uses specific parameter configurations that provide the best trade-off between prediction accuracy and bitrate. This parameter optimization approach improves rate-distortion performance while controlling the number of reference sample combinations through selective parameter usage
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems and methods for improved MRL and TMRL intra prediction, and particularly, the use of unequal offset reference lines in MRL and TMRL to improve intra prediction. In some implementations, the top reference arrays and left reference arrays may have unequal offsets from a target block. Because the top and left reference arrays can have different offsets, there are many more possible combinations for reference samples. The best pair of top and left reference arrays may be signaled with a suitable encoding scheme.