Image Matching via Region Segmentation and Geometric Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image matching technologies rely heavily on local point features, which can be inefficient in determining geometric similarities between images, especially in document image retrieval and copyright violation detection, as they do not effectively utilize geometric relationships between regions of images.
Innovation Solution
The system segments images into regions based on connected components, calculates point descriptors for each region, and identifies geometric relationships by forming a graph with vertices representing region points, allowing for the comparison of geometric constructs between images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If local point features are used for image matching, then the matching process can be performed, but the efficiency and accuracy in determining geometric similarities is insufficient
Solution Approach 1:
The patent segments images into multiple regions based on connected components, transforming the image into a set of discrete region points. This segmentation allows the system to analyze geometric relationships at a higher level (region level) rather than processing individual local point features, thereby improving both the accuracy of geometric similarity determination and the efficiency of the matching process.
Solution Approach 2:
The patent transitions from analyzing local point features in two-dimensional space to analyzing geometric relationships between region points, effectively adding a structural dimension to the feature representation. This dimensional transformation enables more effective capture of geometric similarities while reducing computational complexity.
2Loss of information
If local point features are relied upon exclusively, then the matching algorithm can operate, but geometric relationships between image regions are not effectively utilized
Solution Approach 1:
The patent merges multiple local point features within connected components to form region-level features. By combining information from multiple points into unified region representations, the system preserves geometric relationship information while simplifying the overall feature structure, thus reducing information loss without proportionally increasing complexity.
Solution Approach 2:
The patent creates a simplified geometric construct (graph representation) that copies essential geometric relationships from the original image structure. This abstraction captures the necessary geometric information in a more manageable form, enabling effective utilization of geometric relationships without the full complexity of the original image data.
Data Source
AI summary
Various embodiments disclosed herein relate to systems and methods for image matching involving the segmentation of images into regions and then defining and computing features of these regions to further identify geometrical relationships between the regions.