Image Registration via Mutual Information Maximization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image registration techniques for remote sensing and medical imaging face challenges such as geometric errors, radiometric distortions, and computational inefficiencies, particularly in real-time applications like UAV mapping, due to the need for manual control points, prior knowledge of sensor models, and sensitivity to intensity and contrast variations.
Innovation Solution
The Image Registration by Normalized Mutual Information (IRNMI) method uses wavelet transformation and Fourier-Mellin Transform with phase-only cross correlation and modified Conjugate Gradient Descent optimization to perform sub-pixel geo-spatial registration without requiring prior knowledge of sensor models or radiometric parameters, addressing geometric and radiometric misalignments and achieving robustness against brightness, contrast, and cross-sensor issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual control points and polynomial transformation are used for image registration, then geometric alignment can be achieved, but the process becomes time-consuming and introduces modeling errors due to non-uniform distribution of reliable points
Solution Approach 1:
The patent replaces manual mechanical control point selection with automated mutual information maximization algorithms. The system automatically identifies corresponding features and computes optimal transformation parameters by maximizing mutual information between images, eliminating the need for manual intervention while achieving sub-pixel registration accuracy.
Solution Approach 2:
The patent transforms the registration problem from geometric parameter estimation to radiometric parameter optimization. By formulating registration as a mutual information maximization problem with respect to radiometric parameters, the system achieves robust automatic alignment without requiring manual control points or prior geometric models.
2Measurement precision
If global optimization methods such as simulated annealing and genetic algorithms are used to maximize mutual information, then theoretical optimal alignment can be achieved, but computational cost increases significantly and real-time application becomes infeasible
Solution Approach 1:
The patent segments the optimization process into two distinct stages: a fast coarse registration stage using efficient algorithms to establish initial alignment, followed by a refined fine registration stage that optimizes mutual information locally. This segmentation enables sub-pixel accuracy while maintaining real-time performance by avoiding the computational burden of global optimization methods.
3Extent of automation
If existing mutual information based registration algorithms are used, then automated registration can be achieved, but they break down when source and reference images differ significantly in intensity and contrast variations, cross-sensor phenomenology, and non-linear differences
Solution Approach 1:
The patent fundamentally changes the optimization parameters from geometric transformation parameters to radiometric parameters. By formulating the registration problem as mutual information maximization with respect to radiometric parameters (illumination, atmospheric conditions, sensor characteristics), the system achieves robust automatic registration that is invariant to intensity and contrast variations, cross-sensor differences, and non-linear transformations.
4Extent of automation
If conventional mutual information measure formulation is used for image registration, then automated alignment can be performed, but the solution contains local maximum traps that cause image misalignment instead of reaching global optimum
Solution Approach 1:
The patent divides the registration process into coarse and fine stages, where the coarse stage provides a reliable initial estimate that avoids local maxima, and the fine stage refines the alignment using optimized mutual information maximization. This segmentation ensures the algorithm reaches global optimum by starting from a favorable initial state.
Solution Approach 2:
The patent performs preliminary coarse registration using efficient algorithms to establish an initial alignment before applying fine mutual information optimization. This preliminary action creates a good starting point that guides the optimization away from local maxima and toward the global optimum, ensuring reliable sub-pixel accuracy.
Data Source
Figure 1~2
AI summary
An image registration method includes: providing a reference image and a source image; using a wavelet transformation to produce a transformed reference image and a transformed source image; using the transformed reference image and the transformed source image to estimate affine transform parameters; using the reference image, the source image, and the affine transform estimates to maximize normalized mutual information between the reference image and the source image; and using the normalized mutual information to perform sub-pixel geo-spatial registration of the reference image and the source image to produce an output image. An apparatus that performs the method is also provided.