Motion Estimation Sub-Pixel Search Complexity Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High computational complexity of sub-pixel motion estimation in video compression, particularly in H.265 encoders, due to the need for extensive interpolation, which increases processing demands without significant quality degradation.
Innovation Solution
A method that determines the best full-pixel motion vector and uses this to select reduced sets of half-pixel and quarter-pixel motion vectors, significantly decreasing the number of vectors to be checked, thereby reducing computational complexity while maintaining minimal video quality loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sub-pixel motion estimation with extensive interpolation is performed, then motion estimation accuracy is improved, but computational complexity increases significantly
Solution Approach 1:
The patent segments the motion estimation process into two distinct stages: full-pixel motion estimation followed by sub-pixel motion estimation. By dividing the complex sub-pixel search into a structured two-stage process with candidate selection, the computational burden is reduced while maintaining accuracy. The full-pixel stage provides a coarse search, and the sub-pixel stage refines the result with fewer candidates.
Solution Approach 2:
The patent performs full-pixel motion estimation as a preliminary action before sub-pixel motion estimation. This preliminary full-pixel result is used to generate candidate sub-pixel motion vectors, which narrows down the search space for the subsequent sub-pixel refinement stage, thereby reducing overall computational complexity.
2Measurement precision
If extensive sub-pixel motion vectors are checked, then motion estimation precision is improved, but processing time increases
Solution Approach 1:
The patent segments the motion vector search into full-pixel and sub-pixel stages, with candidate selection acting as a filter between them. This segmentation reduces the number of sub-pixel candidates that need to be evaluated, thereby reducing processing time while maintaining the precision benefits of sub-pixel estimation.
Solution Approach 2:
Instead of checking all possible sub-pixel motion vectors, the patent checks only a selected subset of candidates generated from the full-pixel result. This partial action approach checks sufficient candidates to achieve the desired precision without the excessive computation of examining all possibilities.
3Reliability
If interpolation filters are applied for sub-pixel motion estimation, then video quality is improved, but computational burden increases
Solution Approach 1:
The patent segments the computational workload by first performing motion estimation at full-pixel resolution without heavy interpolation, then applying interpolation filters only for the refined sub-pixel candidate evaluation. This segmentation reduces the overall computational burden while maintaining video quality through selective application of interpolation.
Solution Approach 2:
The patent performs full-pixel motion estimation as a preliminary action before applying interpolation filters for sub-pixel refinement. This preliminary stage establishes a foundation that reduces the number of interpolation operations needed in the subsequent sub-pixel stage, thereby reducing computational burden while preserving video quality.
Data Source
AI summary
A technique of motion estimation in video compression is described, the technique including: determining, in one or more reference frames of a video picture, the best full-pixel motion vector F for a block in a current frame of the video picture, wherein m and n are signed numbers and integer multiples of the distance between two adjacent full-pixels; selecting the best half-pixel motion vector candidates from a set of half-pixel motion vectors based on the best full-pixel motion vector; determining the best half-pixel motion vector H; selecting the best quarter-pixel motion vector candidates from a set of quarter-pixel motion vectors based on the best full-pixel motion vector and the best half-pixel motion vector; determining the best quarter-pixel motion vector Q; and determining the best motion vector for the block as BMV.