Image Alignment via Uniform Feature Point Distribution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image alignment techniques suffer from less accurate results due to uneven distribution of feature points in images, with many points extracted from high contrast regions and few from low contrast regions, leading to inadequate alignment in low contrast areas.
Innovation Solution
An image processing apparatus that divides images into sub-regions and adjusts the feature point extraction threshold to ensure a uniform distribution of feature points across regions, using a division unit and an extraction unit to perform feature point extraction, and optionally generates fake feature points in low contrast areas to enhance alignment accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If feature point extraction uses a fixed threshold based on feature amount, then extraction is simple and fast, but feature points are unevenly distributed with too many in high contrast regions and too few in low contrast regions
Solution Approach 1:
The patent applies local quality by dividing the image into multiple regions and setting different extraction thresholds for each region based on its specific characteristics. High contrast regions use higher thresholds to reduce point density, while low contrast regions use lower thresholds to increase point density, achieving uniform distribution across the entire image.
Solution Approach 2:
The patent changes the extraction threshold parameter dynamically for different regions rather than using a fixed global threshold. By adjusting the threshold parameter according to regional contrast characteristics, the system achieves both efficient extraction and uniform point distribution.
2Productivity
If feature points are extracted only from high contrast regions, then extraction is efficient, but image alignment accuracy deteriorates in low contrast regions
Solution Approach 1:
The patent segments the image into multiple regions and processes each region independently with region-appropriate thresholds. This ensures that low contrast regions receive dedicated attention with lower thresholds, generating sufficient feature points for accurate alignment in those areas while maintaining efficiency in high contrast regions.
Solution Approach 2:
The patent applies different extraction strategies to different regions: high contrast regions use higher thresholds for efficiency, while low contrast regions use lower thresholds to ensure sufficient point generation for accurate alignment, thus achieving both efficiency and accuracy simultaneously.
3Device complexity
If a single extraction threshold is used for the entire image, then the process is simple, but vertical disparity reduction in image alignment is insufficient
Solution Approach 1:
The patent divides the image into multiple regions and applies different thresholds to each segment. This segmentation approach improves vertical disparity reduction accuracy by accounting for regional variations in contrast, while the automated threshold calculation keeps the overall process complexity manageable.
Solution Approach 2:
The patent changes the threshold parameter across different regions based on local contrast characteristics. This parameter variation improves alignment precision and vertical disparity reduction while the thresholds are automatically calculated, preventing excessive complexity in the processing workflow.
Data Source
AI summary
To enhance the accuracy of image alignment performed to reduce a vertical disparity between a first image and a second image that are images of the same object captured from different viewpoints, an image processing apparatus according to an aspect of the present invention includes: a division unit which divides each of the first image and the second image into a plurality of sub-regions; and an extraction unit which performs feature point extraction on each of the sub-regions, wherein the extraction unit performs the feature point extraction in such a manner that a value indicating a degree of variation among the sub-regions in total number of extracted feature points is less than or equal to a predetermined value.