Motion Vector Scaling via Non-Uniform Grid Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Uniform sub-pixel motion vector grids may not be optimal for given filter restrictions and power spectral density of reference blocks in high-efficiency video coding, leading to inaccurate motion vector scaling.
Innovation Solution
A method that determines a scaled motion vector for a block by mapping motion vector values from a non-uniform grid to a higher accuracy uniform grid, scaling them, and then mapping back to the non-uniform grid for temporal prediction, allowing for more accurate sub-pixel position representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a uniform sub-pixel motion vector grid is used, then the motion vector grid structure is simple and easy to implement, but it does not align optimally with filter restrictions and power spectral density characteristics, leading to inaccurate motion vector scaling
Solution Approach 1:
The patent changes the parameter of sub-pixel phase offsets from uniform distribution to non-uniform distribution. Specifically, it sets the first sub-pixel phase offset to 1/8 pixel, the second to 1/2 pixel, and the third to 7/8 pixel, which are optimally positioned according to filter restrictions and power spectral density characteristics. This parameter change improves motion vector scaling accuracy while maintaining grid structure simplicity.
2Ease of operation
If sub-pixel positions are uniformly distributed, then the interpolation process is straightforward, but it fails to account for object motion that does not align with integer pixel spacing, reducing prediction accuracy
Solution Approach 1:
The patent applies local quality by positioning sub-pixel phases at specific locations (1/8, 1/2, and 7/8 pixel offsets) rather than uniformly distributing them. These locally optimized positions correspond to regions of higher energy in the power spectral density and align better with actual object motion patterns, thereby improving sub-pixel position accuracy while keeping the interpolation process manageable.
3Productivity
If motion vectors are scaled on a uniform grid, then the scaling calculation is simple, but it produces errors when mapping between different temporal distances, reducing video compression efficiency
Solution Approach 1:
The patent performs preliminary action by pre-defining optimal sub-pixel phase positions (1/8, 1/2, 7/8 pixel offsets) based on filter restrictions and power spectral density analysis. This preliminary optimization of the motion vector grid structure ensures that subsequent motion vector scaling operations between different temporal distances are more accurate, reducing scaling errors and improving video compression efficiency.
Data Source
AI summary
In one embodiment, a method determines a scaled motion vector for a first block. A motion vector for a second block is determined where the motion vector is on a non-uniform motion vector grid. The method then maps the motion vector for the second block to a higher accuracy uniform motion vector grid that is of a higher accuracy than the non-uniform motion vector grid and scales the motion vector for the second block on the higher accuracy motion vector grid. The scaled motion vector is mapped on the higher accuracy motion vector grid to the non-uniform motion vector grid. The scaled motion vector on the non-uniform motion vector grid is associated with the first block for a temporal prediction process.