Image Registration via Optimization Over Disjoint Regions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image registration for large collections of images with numerous features is complex and prone to errors due to distortion and parallax, especially when features or regions in the imagery change over time or exhibit inconsistent motion.
Innovation Solution
An image registration algorithm that employs an optimization metric computed over a selected, disjoint set of separately normalized image regions, allowing the exclusion of regions with inconsistent motion and facilitating the selection of subregions for accurate registration, using techniques like Nelder-Mead simplex optimization and optical distortion correction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If optimization-based registration is used to minimize difference metric over entire image, then registration accuracy is improved, but computational complexity increases and error from inconsistent motion regions propagates
Solution Approach 1:
The patent divides the image into multiple disjoint regions and computes the optimization metric separately for each region rather than over the entire image. This segmentation allows the algorithm to exclude regions with inconsistent motion (such as moving objects) from the registration calculation, preventing error propagation while reducing computational complexity compared to processing the full image.
Solution Approach 2:
The patent applies different treatment to different regions of the image by selecting disjoint regions that exhibit consistent motion characteristics. Each region is separately normalized and evaluated, allowing the algorithm to weight reliable regions more heavily while excluding problematic regions, thereby improving overall registration accuracy without uniform computational expense.
2Productivity
If feature-based registration is used to reduce computational complexity, then processing speed is improved, but accuracy deteriorates due to distortion and parallax
Solution Approach 1:
The patent extracts and excludes specific regions with inconsistent motion or parallax errors from the registration calculation. By identifying and removing these problematic regions from the optimization metric computation, the algorithm achieves accuracy comparable to full-image optimization while maintaining the computational efficiency of selective processing.
3Reliability
If entire image is used for optimization metric computation, then comprehensive registration is achieved, but error from parallax and inconsistent motion affects overall accuracy
Solution Approach 1:
The patent segments the image into multiple disjoint regions and computes the optimization metric for each region separately. This allows the algorithm to identify and exclude regions with inconsistent motion or parallax errors, ensuring that the registration is based only on reliable regions with consistent spatial relationships, thereby improving both consistency and accuracy.
Solution Approach 2:
The patent implements a feedback mechanism where the optimization metric is evaluated for each disjoint region, and regions exhibiting inconsistent motion or large errors are identified and excluded from the final registration calculation. This feedback loop ensures that only reliable regions contribute to the final transform parameters.
Data Source
AI summary
Technologies pertaining to registering a target image with a base image are described. In a general embodiment, the base image is selected from a set of images, and the target image is an image in the set of images that is to be registered to the base image. A set of disjoint regions of the target image is selected, and a transform to be applied to the target image is computed based on the optimization of a metric over the selected set of disjoint regions. The transform is applied to the target image so as to register the target image with the base image.