Gradient Domain Image Registration via Total Variation Minimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image registration methods face challenges in aligning images with significant intensity variations, such as those from satellite and aerial images, due to intrinsic and extrinsic factors, leading to misalignment and high computational complexity, especially when dealing with intensity distortions and artifacts.
Innovation Solution
The method employs image registration in the gradient domain using an energy minimization function based on total variation, which matches image edges and applies a gradient descent algorithm to reduce differential total variation, thereby creating a sparse composite image without requiring regularization parameters and reducing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If intensity-based registration methods are used to handle images with intensity variations, then alignment accuracy may be maintained, but robustness to intensity distortions deteriorates
Solution Approach 1:
The patent introduces gradient magnitude and gradient direction as intermediary representations instead of using raw intensity values directly. By transforming the images into the gradient domain, the method mediates between intensity-based alignment and robustness to intensity distortions, allowing alignment based on edge structures that are invariant to intensity variations and artifacts.
2Reliability
If simultaneous registration and intensity correction methods are applied to handle intensity artifacts, then alignment robustness improves, but computational complexity increases
Solution Approach 1:
The patent extracts only the gradient magnitude and gradient direction components from the original images, discarding the problematic intensity information. This extraction approach separates the useful structural information (edges) from the harmful intensity variations, achieving robustness without the computational burden of simultaneous intensity correction.
Solution Approach 2:
The patent replaces the complex mechanical system of simultaneous registration and intensity correction with a simpler gradient-domain matching approach. By substituting the intensity-based similarity measure with gradient-based comparison, the method achieves similar robustness with significantly reduced computational complexity.
3Measurement precision
If regularization parameters are used in residual complexity methods to handle intensity distortions, then measurement accuracy improves, but device complexity increases
Solution Approach 1:
The gradient-domain matching method is self-sufficient and does not require external regularization parameters or additional tuning variables. The gradient magnitude and direction inherently provide the necessary constraints for robust alignment without needing extra parameters, making the system self-service and simpler.
Data Source
AI summary
A method, apparatus and computer program product are provided for image registration in the gradient domain. A method is provided including receiving a first image and second image; and registering the first and second images in a gradient domain. The registration of the first and second images in the gradient domain includes applying an energy minimization function based on total variation.