Hierarchical Motion Vector Refinement for Video Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hierarchical motion estimation in video processing often results in erroneous motion vectors due to ambiguity at lower resolutions and object boundary issues, leading to artifacts like dragging effects, and existing solutions either degrade compression efficiency or increase computational complexity.
Innovation Solution
A method that refines motion vectors by using initial and neighboring motion vectors from external blocks to perform small search-range refinement searches, replacing current motion vectors with better estimates, and propagating refined vectors to subsequent layers, thereby reducing computational complexity and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional search with large search range is used, then high motion and high temporal separation are captured, but compression efficiency is degraded
Solution Approach 1:
The patent segments the motion estimation process into hierarchical layers, dividing the large search space into multiple resolution levels. Each layer performs motion estimation on a subset of the search space, with results propagated to subsequent layers. This segmentation allows accurate motion vector estimation without requiring exhaustive full-resolution searches, thereby maintaining compression efficiency while capturing high motion and temporal separation.
2Device complexity
If hierarchical motion estimation is used, then computational complexity is reduced, but erroneous motion vectors occur due to ambiguity at lower resolutions
Solution Approach 1:
The patent implements a nested hierarchical structure where motion estimation results from coarser layers are embedded and propagated to finer layers. Each layer's motion vectors serve as initial estimates for the next layer, creating a nested refinement process. This nesting allows the system to benefit from reduced computational complexity at higher layers while progressively improving accuracy at lower layers through refinement searches.
Solution Approach 2:
The patent performs preliminary motion estimation at coarser resolutions before conducting detailed searches at finer resolutions. The motion vectors obtained from higher layers serve as preliminary estimates that guide and constrain subsequent searches at lower layers. This preliminary action reduces the search space for finer layers, preventing erroneous vectors while maintaining computational efficiency.
3Device complexity
If motion vector propagation is performed from higher layers, then computational complexity is reduced, but erroneous vectors propagate to lower layers
Solution Approach 1:
The patent incorporates feedback mechanisms where motion vectors from higher layers are used as initial estimates for lower layers, but the results from lower layers are used to refine and correct the propagated vectors. The refinement process at each layer provides feedback that corrects errors introduced during propagation, ensuring that erroneous vectors do not persist to lower resolutions while maintaining the computational benefits of hierarchical propagation.
Data Source
AI summary
A method of motion estimation (ME) refinement. The method generally includes the steps of (A) generating an initial motion vector (MV) by conducting a first ME on an initial block in a picture, the initial block covering an initial area of the picture, (B) generating a current MV by conducting a second ME on a current block in the picture, (i) the current block covering a subset of the initial area and (ii) the second ME being seeded by the initial MV, (C) generating at least one additional MV by conducting at least one third ME on the current block, the at least one third ME being seeded respectively by at least one neighboring MV spatially adjacent to the current MV and (D) generating a refined MV of the current block by storing in a memory a best among the current MV and the additional MV.