Fractional-Pel Prediction on Non-Uniform Sampling Grids
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video encoding and decoding methods face high complexity and inaccuracies due to multiple motion compensated predictions, as each prediction does not benefit from knowledge of samples from other reference blocks, leading to less accurate combined predictions.
Innovation Solution
Implementing a method where reference samples are projected onto a fractional-pel sampling grid and interpolated using non-uniform positioning, followed by a two-dimensional interpolation process to generate a prediction block, which includes determining global weights for reference blocks and applying them to generate a weighted sum.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple motion compensated predictions are performed using multiple reference blocks, then prediction accuracy can be improved, but computational complexity increases linearly
Solution Approach 1:
The patent merges multiple motion compensated predictions into a single combined prediction operation. Instead of separately computing each prediction and then combining results, the method projects reference samples from multiple reference blocks onto a unified fractional-pel sampling grid and performs a single interpolation operation to generate the final prediction block, thereby reducing computational complexity while maintaining prediction accuracy
Solution Approach 2:
The patent introduces a new dimensional approach by projecting reference samples onto a fractional-pel sampling grid. This transformation from separate reference block coordinates to a unified fractional-pel grid enables efficient combined interpolation operations that reduce computational complexity while preserving the benefits of multiple reference blocks
2Device complexity
If each motion compensated prediction is performed independently from its own reference block, then computational simplicity is maintained, but prediction accuracy deteriorates due to lack of knowledge from other reference blocks
Solution Approach 1:
The patent introduces a fractional-pel sampling grid as an intermediary structure that mediates between multiple reference blocks. Reference samples from different reference blocks are projected onto this common grid, allowing information from all reference blocks to be integrated into a single prediction operation, thereby improving accuracy while maintaining computational efficiency
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems and methods for video encoding and/or video decoding are provided. An encoder and/or decoder performs a method including receiving a plurality of reference blocks from one or more reference pictures for decoding a current block. The plurality of reference blocks are identified by one or more two-dimensional motion vectors including a sum of a full-pel motion vector and a fractional-pel motion vector. A plurality of reference samples of at least one reference block are projected onto a fractional-pel sampling grid associated with the current block. The plurality of reference samples may be positioned non-uniformly onto the fractional-pel sampling grid. A prediction block is generated by interpolating one or more samples from the projected plurality of reference samples. The method reconstructs the current block using the generated prediction block. The method performs a one-step two dimensional interpolation and/or a two-step interpolation including two interpolations, viz. horizontal interpolation and vertical interpolation.