Implicit Masked Blending with HLS for Boundary-Aware Video Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding systems inaccurately predict block sections when reference blocks extend beyond picture boundaries due to padding, leading to less accurate video reconstruction and increased computational costs.
Innovation Solution
Implement an implicit masked blending mode using high-level syntax flags to selectively apply weighted averaging or non-averaged blending based on reference block positions relative to picture boundaries, ensuring accurate reconstruction while optimizing computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If reference blocks are padded when extending beyond picture boundaries, then the prediction process can be completed, but the prediction accuracy deteriorates due to use of non-real pixel values
Solution Approach 1:
The patent extracts and removes the problematic padded regions from reference blocks that extend beyond picture boundaries. By identifying and excluding these non-real pixel value regions, the system prevents them from degrading prediction accuracy while retaining the useful real pixel data within valid boundaries.
Solution Approach 2:
The patent applies different quality treatments to different regions of reference blocks. Regions within picture boundaries receive full weight as they contain real pixel data, while regions beyond boundaries are either excluded or given reduced weight since they contain padded non-real values. This local differentiation optimizes overall prediction accuracy.
2Manufacturing precision
If non-averaged weighting is applied to reference blocks, then reconstruction accuracy improves by weighting real pixel data more heavily, but computational cost increases
Solution Approach 1:
The patent implements partial weighting adjustment rather than full complex weighting schemes. By applying simplified weight factors that only partially adjust for boundary conditions (e.g., linear gradients or fixed weight ratios), the system achieves meaningful accuracy improvement without the full computational burden of complex non-averaged weighting algorithms.
Solution Approach 2:
The patent changes the weighting parameters from uniform averaging to position-dependent weights. By introducing simple parametric weight functions that vary based on distance from picture boundaries or block position, the system achieves better reconstruction accuracy through parameter adjustment rather than complex computational processes.
3Manufacturing precision
If implicit masked blending mode is implemented, then prediction accuracy improves by using real pixel data, but device complexity increases
Solution Approach 1:
The patent performs preliminary identification and masking of boundary regions before the main prediction process. By pre-computing mask patterns or weight maps that indicate valid versus invalid reference regions, the system eliminates the need for complex runtime decisions about which pixels to use, simplifying the actual prediction execution.
Solution Approach 2:
The patent introduces an intermediate masking layer or weight map that mediates between the reference blocks and the prediction process. This intermediary structure simplifies the interaction by pre-processing the reference data to highlight valid regions, making the subsequent prediction operation simpler and more accurate without directly complicating the core prediction algorithm.
Data Source
AI summary
An example method of video encoding includes receiving video data comprising first block. The method also includes determining that the first block is to be encoded using first and second reference blocks and signaling a high-level syntax (HLS) flag in a video bitstream that indicates whether an implicit masked blending mode is enabled for the first block. When the HLS flag having a first value, encoding the first block using a first set of weighting factors for reference values from the first reference block and the second reference block. When (i) the HLS flag having a second value and (ii) at least one of the first reference block and the second reference block is at least partially outside of a corresponding reference boundary, encoding the first block using a second set of weighting factors for the reference values from the first reference block and the second reference block.