Image Registration via Spatial Frequency Distribution Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image registration methods are adversely affected by noise, such as low-light conditions and blur, leading to suboptimal results in terms of robustness and computational complexity, especially in feature-based methods and phase correlation techniques.
Innovation Solution
A method that adjusts the distribution of spatial frequencies in the correlation between images using a frequency domain filter, which reduces high-frequency noise while preserving valid high-frequency contents, and incorporates imaging parameters like brightness, contrast, and sensitivity to enhance noise robustness and computational efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing denoise methods are applied prior to registration, then noise is reduced, but computational complexity increases and robustness remains insufficient
Solution Approach 1:
The patent applies preliminary denoising action by adjusting the distribution of spatial frequencies in the correlation between images before registration. A frequency domain filter is used to reduce high-frequency noise while preserving valid high-frequency contents, preparing the images for more effective registration without requiring complex post-processing denoising methods
Solution Approach 2:
The patent changes parameters in the frequency domain by adjusting the distribution of spatial frequencies. Imaging parameters such as brightness, contrast, and sensitivity are incorporated to adaptively control the filtering process, reducing noise while maintaining registration accuracy without increasing computational complexity
2Measurement precision
If feature-based methods are used for registration, then registration accuracy is improved, but computational complexity and sensitivity to noise increase
Solution Approach 1:
The patent extracts only the necessary high-frequency contents that are valid for registration while removing harmful high-frequency noise. By selectively preserving useful frequency components and filtering out noise, the method achieves accurate registration without requiring complex feature detection and matching algorithms
Solution Approach 2:
The patent introduces a frequency domain filter as an intermediary between the noisy images and the registration process. This filter mediates by adjusting the distribution of spatial frequencies, allowing simple correlation-based registration to achieve accuracy comparable to complex feature-based methods while maintaining lower computational complexity
3Productivity
If phase correlation techniques are applied, then registration speed is improved, but robustness against noise and blur deteriorates
Solution Approach 1:
The patent applies preliminary frequency domain filtering to the images before phase correlation registration. By adjusting the distribution of spatial frequencies to reduce noise and blur effects in advance, the simple and fast phase correlation technique becomes robust against noisy and blurred images while maintaining its computational efficiency and registration speed
Data Source
AI summary
A method of registering images by adjusting a distribution of spatial frequencies of correlation between spatial frequency functions of images, a number of images being N, where N being an integer equal to or larger than two, and determining at least one shift between the images using the frequency distribution adjusted correlation.