Multigraph Optimization for Image Stitching Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image stitching methods face difficulties in finding correct transformations and point correspondences when overlapping regions contain repetitive patterns or lack detail, leading to incorrect matches and visible stitching artifacts.
Innovation Solution
A method involving the formation of a multigraph with nodes representing transformations from a composite image to individual images, using optimization to solve edge weights and nodal transformations, and applying these transformations to create a seamless composite image, while incorporating dummy edges to account for non-matches and pruning the graph to select the most plausible transformations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional pairwise matching method is used to determine transformation between image pairs, then the stitching process is simple and fast, but the method fails to find correct transformations when overlapping regions contain repetitive patterns or lack detail
Solution Approach 1:
The patent segments the transformation determination process into multiple candidate transformations instead of relying on a single pairwise match. By generating and evaluating multiple candidate transformations with associated weights, the system can select the most plausible transformation even in challenging overlapping regions with repetitive patterns or low detail.
Solution Approach 2:
The patent introduces a weight dimension to the transformation selection process. Each candidate transformation is assigned a weight indicating its plausibility, transforming the selection from a simple binary choice to a weighted optimization problem. This additional dimension allows the system to evaluate and compare multiple transformations simultaneously.
2Measurement precision
If multiple candidate transformations are formed between image pairs, then the accuracy of transformation selection improves, but the computational complexity and processing time increase
Solution Approach 1:
The patent generates multiple candidate transformations (excessive action) but only processes them to the extent necessary for making a reliable selection. By using weight-based filtering and optimization, the system processes more candidate transformations than the conventional single-match approach while maintaining efficiency through selective evaluation.
Solution Approach 2:
The patent changes the parameter of transformation selection from finding a single best match to evaluating multiple candidates with weight parameters. This parameter change allows the system to balance precision and speed by adjusting the number of candidate transformations evaluated and the threshold for selection.
3Reliability
If dummy edges are added to the multigraph to account for non-matches, then the robustness against false matches improves, but the graph complexity and optimization problem size increase
Solution Approach 1:
The patent introduces dummy edges as intermediary elements in the multigraph that represent the possibility of no valid transformation between image pairs. These dummy edges act as mediators that allow the optimization algorithm to gracefully handle cases where no good match exists, preventing false matches by providing a fallback option with appropriate weight.
Solution Approach 2:
The patent prepares for potential false matches by beforehand including dummy edges in the multigraph construction. This prior cushioning ensures that when the optimization process encounters ambiguous or non-matching image pairs, the dummy edges provide a pre-prepared safe option, preventing the system from selecting incorrect transformations.
Data Source
AI summary
There is provided a method comprising receiving at least three images, wherein the images form a number of partially overlapping image pairs, wherein the number of partially overlapping image pairs is at least a number of the received images; forming one or more candidate transformations between the images of the image pairs; constructing a multigraph comprising nodes representing nodal transformations from a composite image to the images of the at least three images, and edges between the nodes, the edges representing the one or more candidate transformations; solving edge weights and the nodal transformations using an optimization problem, wherein the edge weights indicate plausibility of the one or more candidate transformations; and applying the solved nodal transformations in forming the composite image of the at least three images.