Multispectral Image Target Detection via Local Window Contrast Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In terrestrial environments with complex landscapes and camouflaged targets, existing multispectral image analysis methods fail to reliably detect intruding elements due to low contrast in images, even when using Fisher projection optimization across the entire image field.
Innovation Solution
A method that optimizes contrast within localized windows of the image matrix, using Fisher projection to enhance target detection by reducing dilution effects from non-correlated spectral contributions, and presents the optimized image to the operator for improved comfort and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Fisher projection optimization is applied across the entire image field, then the contrast of the resulting image is at least equal to that of each separate spectral image, but the contrast is diluted by non-correlated spectral contributions from the entire field
Solution Approach 1:
The patent divides the image field into multiple local windows and applies Fisher projection optimization independently to each window. This segmentation prevents dilution of contrast by non-correlated spectral contributions from distant regions, while maintaining the benefits of multispectral analysis within each local area where spectral correlations are more meaningful.
Solution Approach 2:
The patent implements local quality by optimizing contrast separately for each local window rather than applying a global optimization. This allows each region to have its own optimized projection direction tailored to its specific spectral characteristics, improving detection reliability in heterogeneous environments with varying background compositions.
2Productivity
If the covariance matrix is estimated for the entire field of the image, then the optimal direction of projection can be determined, but this global estimation includes non-correlated spectral contributions that reduce target contrast
Solution Approach 1:
The patent segments the image field into local windows and estimates the covariance matrix separately for each window. This local covariance estimation excludes non-correlated spectral contributions from distant regions, resulting in more precise contrast optimization for each local area while maintaining overall detection efficiency through parallel processing of multiple windows.
3Adaptability or versatility
If separate spectral images are analyzed, then each spectral band can be examined individually, but the target may not be detected distinctly enough given the observation time
Solution Approach 1:
The patent merges information from multiple spectral bands by applying Fisher projection to combine the spectral intensity values into a single optimized image. This combining approach maintains the flexibility of individual spectral band analysis while improving detection reliability through the synergistic integration of multispectral information with enhanced contrast.
Data Source
Figure 1~4b
Figure 3a~3d
Figure 5~6
AI summary
A search for a target in a multispectral image is made more efficient and more user-friendly by combining a contrast optimization which is performed locally, with a presentation of a detection image which extends over the entire field of observation (10). The contrast is optimized inside a window (2) of reduced size relative to an image matrix (1) corresponding to the entire field of observation. This window maybe moved in conjunction with the direction of observation (D), or it maybe selected at will in the image matrix. The detection image maybe renewed for each window used, or it maybe shared by several windows.