Off-band Image Resolution Enhancement via High-Pass Detail Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image enhancement methods for remote sensing systems struggle to effectively increase the spatial resolution of low-resolution imagery without spectral overlap and lack control over the sharpening process, particularly in multi-spectral and mixed modality imagery.
Innovation Solution
A method that involves increasing the sampling rate of a first image to form an interpolated image, processing a second image through a high pass filter to extract detail, and merging this detail with the interpolated image using segment-specific prediction coefficients, allowing for enhancement across non-overlapping spectral bands and modalities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If conventional pansharpening methods are used to enhance low-resolution imagery, then spatial resolution is improved, but the method requires spectral overlap between bands which limits applicability
Solution Approach 1:
The image is segmented into multiple regions based on spectral and spatial characteristics. Different prediction models are applied to different segments, allowing the method to handle diverse spectral relationships including non-overlapping bands. This segmentation enables flexible adaptation to various spectral configurations while maintaining enhancement quality.
Solution Approach 2:
The method dynamically adjusts prediction coefficients and model parameters based on the spectral relationship between bands. By changing parameters adaptively rather than requiring fixed spectral overlap conditions, the method can effectively enhance images across different spectral configurations including non-overlapping bands.
2Manufacturing precision
If conventional pansharpening methods are used, then spatial resolution is enhanced, but control over the sharpening amount is lost
Solution Approach 1:
The method incorporates feedback mechanisms where prediction coefficients are optimized based on the relationship between high-resolution reference bands and low-resolution target bands. This feedback loop allows dynamic control over the enhancement amount, enabling users to adjust sharpening intensity while maintaining quality.
Solution Approach 2:
The enhancement process is made dynamic through adjustable prediction coefficients and multiple model options. Users can control the degree of sharpening by selecting different models and adjusting parameters, transforming a static fixed-strength process into a dynamic controllable one.
3Reliability
If spectral bandwidth is increased to improve signal-to-noise ratio, then SNR is improved, but spatial resolution decreases
Solution Approach 1:
The method merges information from multiple spectral bands including both broad-band high-SNR images and narrow-band high-resolution images. By combining these complementary data sources through predictive modeling, the method achieves both high signal-to-noise ratio and high spatial resolution in the enhanced output.
Solution Approach 2:
High-resolution bands serve as intermediaries to transfer spatial detail information to low-resolution bands. This intermediary mechanism allows the transfer of fine spatial structures from high-resolution narrow-band images to enhance the corresponding regions in low-resolution broad-band images, achieving both SNR and resolution improvements.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method (100) of enhancing an image includes increasing sampling rate of a first image (102) to a target sampling rate to form an interpolated image (146). The method also includes processing a second image (104) through a high pass filter (114) to form a high pass features image (116), wherein the second image (104) is at the target sampling rate. The method also includes extracting detail (118) from the high pass features image (116) relevant to the first image (102), merging (120) the detail from the high pass features image (116) with the interpolated image (146) to form a prediction image (106) at the target sampling rate, and outputting the prediction image (106).