Image Registration Using Gradient Segmentation for Real-Time Stitching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image registration methods, such as the correlation coefficient method, are not suitable for real-time applications and produce inaccurate results when the content difference between floating and reference images is significant, especially in high-speed microscope scanning.
Innovation Solution
An image registration method that involves performing region segmentation on gradient images using a region growing algorithm and determining candidate pixels with extreme correlation coefficients based on gradient thresholds, allowing for accurate registration without iterative calculations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the correlation coefficient method is used to avoid iterative calculations, then registration speed is improved, but registration accuracy deteriorates when content difference between images is significant
Solution Approach 1:
The patent segments the image into multiple regions based on gradient magnitude thresholds. By dividing the image into distinct regions with different gradient characteristics, the method can selectively apply correlation analysis to relevant regions while ignoring regions with significant content differences, thus maintaining both speed and accuracy.
Solution Approach 2:
The patent applies different processing strategies to different regions of the image based on their gradient characteristics. Regions with high gradient magnitude are treated differently from regions with low gradient magnitude, allowing the method to adapt to local content differences and maintain registration accuracy without sacrificing overall speed.
2Measurement precision
If iterative methods are used to optimize cost function, then registration accuracy is improved, but calculation time increases making real-time applications difficult
Solution Approach 1:
The patent replaces the iterative optimization process (mechanical system) with a direct correlation-based method combined with gradient threshold segmentation. This substitution eliminates the need for repeated calculations while maintaining accuracy by using the gradient information to guide the correlation analysis directly to the correct registration position.
3Productivity
If correlation coefficient method is used for fast registration, then calculation time is reduced, but reliability deteriorates when probe moving speed is high
Solution Approach 1:
The patent dynamically adjusts the gradient threshold and region segmentation based on the image content and movement characteristics. By making the segmentation criteria adaptive rather than fixed, the method can maintain reliability under varying probe speeds while preserving the speed advantage of the correlation coefficient method.
Data Source
Figure 1
Figure 2A~3
Figure 4~5
AI summary
Embodiments of the present invention provide an image registration method and apparatus, and an image stitching method and device. The image registration method includes: calculating a correlation coefficient image between a reference image and a floating image; calculating a gradient image of the correlation coefficient image; determining a pixel with extreme correlation coefficient in the correlation coefficient image according to a pixel with extreme gradient in the gradient image; and registering the reference image and the floating image according to the pixel with extreme correlation coefficient. The above image registration method and apparatus have high universality, require a small amount of calculation, and can ensure real-time operation and obtain a more accurate registration result. The image stitching method and device are realized based on the image registration method, thereby improving the speed and accuracy.