Sparsity-Driven Image Registration for Robust Multi-Modal Fusion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image registration methods face challenges in achieving robust and accurate registration of multiple rigid transformed images, particularly in fusing high spatial and spectral resolution images, often resulting in mis-registration and spectral distortion.
Innovation Solution
A sparsity-driven image registration method that decomposes registration parameter matrices into low-rank and sparse components, using mutual information-based approaches to generate and update registration parameters, thereby improving the robustness and accuracy of image registration.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image registration methods are used to register multiple images, then the registration process can be completed, but the registration accuracy deteriorates and spectral distortion occurs
Solution Approach 1:
The patent segments the registration parameter matrix into two distinct components: a low-rank matrix representing the true registration parameters and a sparse matrix representing outliers and errors. This segmentation allows the method to separately handle and correct erroneous registration parameters while preserving accurate ones, thereby improving registration accuracy and robustness simultaneously.
Solution Approach 2:
The patent introduces an intermediary mathematical model (low-rank plus sparse decomposition) that acts as a mediator between the observed registration parameters and the true registration parameters. This intermediary model enables the separation of signal from noise and outliers, resolving the contradiction between accuracy and robustness.
2Productivity
If multiple images are registered using conventional methods, then all images can be processed, but spectral distortion increases and fusion performance deteriorates
Solution Approach 1:
The patent extracts and removes the sparse outlier components from the registration parameter matrix. By taking out the erroneous parameters (outliers) from the overall registration process, the method prevents these errors from causing spectral distortion in the fused images, while maintaining the ability to process multiple images efficiently.
3Loss of information
If registration parameters are estimated for all image pairs, then complete registration information is obtained, but outlier errors increase and registration robustness decreases
Solution Approach 1:
The patent applies local quality by treating different elements of the registration parameter matrix differently. Accurate registration parameters are preserved with their full information content, while outlier parameters are identified and corrected through the sparse decomposition. This local differentiation maintains information completeness while improving robustness against outliers.
Data Source
AI summary
Systems and methods for multiple image registration of images of a scene or an object. Receiving image data, the image data includes images collected from different measurements of a single modality or multiple modalities, either at different rotation angles, horizontal shifts, or vertical shifts, of the scene or the object. Estimating registration parameters, using pairs of images, each pair of images includes a reference image and a floating image. Generating parameter matrices corresponding to registration parameters using an image registration process for all pairs of images. Decomposing each parameter matrix into a low-rank matrix of updated registration parameters and a sparse matrix corresponding to the registration parameter errors for each low-rank matrix, to obtain updated registration parameters for robust registration. Using the updated registration parameters to form a transformation matrix to register the images with at least one reference image, resulting in robust registration of the images.