Motion Estimation for Periodic Patterns Using Global Vector Penalties
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion estimation methods for periodic patterns in video compression often result in incorrect motion vectors due to high similarity among blocks, leading to poor image quality and blurring or serrated edges, as they fail to accurately determine motion vectors when candidate vectors have small Sum of Absolute Difference (SAD) values.
Innovation Solution
A method and apparatus that determine a global motion vector for regions within a frame, generate candidate motion vectors for target blocks, calculate penalty values based on differences with global motion vectors, and compute weighted similarity values to accurately select the correct motion vector, thereby reducing the likelihood of incorrect motion vector selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If block matching motion estimation is used for periodic patterns, then video compression is achieved, but incorrect motion vectors are selected due to high similarity among blocks
Solution Approach 1:
The patent changes the evaluation parameter from simple SAD (Sum of Absolute Difference) to a composite metric that incorporates penalty values. This penalty value is calculated based on the difference between candidate motion vectors and global motion vectors, effectively transforming the selection criterion to account for periodic pattern characteristics and resolve the ambiguity caused by high block similarity.
Solution Approach 2:
The patent introduces global motion vectors as an intermediary element. These global motion vectors are calculated from multiple periodic blocks and serve as a reference to evaluate candidate motion vectors. This intermediary helps distinguish the true motion vector from false candidates that have similarly low SAD values, thereby improving motion vector accuracy in periodic patterns.
2Productivity
If candidate motion vectors with small SAD values are selected, then matching efficiency is improved, but image quality deteriorates due to incorrect motion vector selection
Solution Approach 1:
The patent modifies the selection parameter by introducing a penalty component to the existing SAD metric. The new selection criterion becomes SAD plus penalty value, where the penalty value is derived from the discrepancy between candidate motion vectors and global motion vectors. This parameter change maintains computational efficiency while significantly improving the reliability of motion vector selection for periodic patterns.
Solution Approach 2:
The patent implements a feedback mechanism where global motion vectors, derived from multiple periodic blocks, are used to evaluate and adjust the selection of individual block motion vectors. This feedback loop ensures that selected motion vectors are consistent with the overall periodic pattern motion, preventing incorrect selections that would degrade image quality.
3Device complexity
If zero motion vector is selected for candidate blocks, then computational complexity is reduced, but motion estimation accuracy deteriorates for periodic patterns
Solution Approach 1:
The patent changes the evaluation parameter to include a penalty term that prevents premature selection of zero motion vectors. Even when SAD values are low (suggesting no motion), the penalty value based on global motion vector differences ensures that actual periodic pattern motion is detected, maintaining accuracy without significantly increasing computational complexity.
Data Source
AI summary
A motion estimation method for blocks of a periodic pattern is provided, which includes determining a global motion vector corresponding to a region according to motion vectors of periodic blocks in the region; generating candidate motion vectors of a target periodic block to be encoded in a second frame; for each candidate motion vector, determining a penalty value based on at least one difference between the candidate motion vector and at least one global motion vector corresponding to at least a relative region in the first frame; for each candidate motion vector, calculating a weighted similarity value based on an original similarity value between the target periodic block of the second frame and a reference block corresponding to the candidate motion vector of the first frame, and the penalty value; and determining a motion vector of the target periodic block according to weighted similarity values of the candidate motion vectors.