Pixel Data Filling in Remote Sensing Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques fail to accurately fill in or replace missing pixel data in aerial and satellite images, often resulting in poorly corrected images due to mechanical failures or obstructions like clouds and shadows.
Innovation Solution
A method that aligns and re-samples source and target images to match resolution, classifies pixel data, and applies scaling factors to fill in or replace missing data from a source image captured at a different time or with a different imaging system, ensuring accurate representation of the terrain.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pixel duplication techniques are used to fill in missing pixel data, then the correction process is simple and fast, but the image quality and accuracy of the corrected regions deteriorate
Solution Approach 1:
The patent segments the image processing task by separating the correction process into multiple stages: identifying missing pixel regions, selecting source pixels from surrounding areas, and applying weighted averaging rather than simple duplication. This segmentation allows each stage to be optimized independently, improving overall accuracy while maintaining reasonable processing speed.
Solution Approach 2:
The patent applies local quality by using spatially varying correction methods. Instead of a uniform approach, it selects source pixels from local neighborhoods around each missing pixel and applies weighted averaging where weights depend on the spatial relationship and similarity of pixels. This ensures that each region is corrected according to its local characteristics, improving accuracy without requiring globally complex processing.
2Measurement precision
If complex image processing techniques are applied to fill in missing pixel data, then image accuracy improves, but the processing complexity and computational resources required increase
Solution Approach 1:
The patent applies partial action by focusing computational resources only on regions with missing pixel data rather than processing the entire image. It identifies and corrects only the affected areas using targeted pixel selection and weighted averaging, achieving high accuracy in corrected regions while avoiding unnecessary computation in already-valid areas, thus balancing accuracy with processing complexity.
Solution Approach 2:
The patent uses copying by selecting source pixels from existing valid regions and copying their values (with weighting) to fill missing pixels. Rather than generating new data through complex reconstruction algorithms, it intelligently copies from proven good pixels, simplifying the processing while maintaining accuracy through careful selection and weighting of source pixels.
3Ease of manufacture
If pixel data from neighboring pixels are assumed to be similar, then the correction method is simple to implement, but the corrected image quality deteriorates due to poor approximation
Solution Approach 1:
The patent changes the parameter of pixel similarity by introducing weighted averaging instead of assuming uniform similarity. It calculates weights based on spatial distance and spectral similarity, transforming the simple assumption into a refined parameter-based approach. This maintains implementation feasibility while dramatically improving approximation quality through mathematically grounded weighting schemes.
Solution Approach 2:
The patent incorporates feedback by evaluating the similarity between source pixels and target pixels, then using this evaluation to adjust the weighting in the averaging process. Pixels that are more similar contribute more to the correction, while less similar pixels contribute less. This feedback mechanism ensures high approximation quality while keeping the implementation straightforward through iterative refinement.
Data Source
AI summary
A system for filling in and/or replacing pixel data in a target image uses pixel data from a source image. In one embodiment, the pixel data in the source image are classified and boundaries of local class areas or groups of similarly classified pixels are determined. The pixel data in the local class areas are compared to determine one or more scaling factors. The missing pixel data or data to be replaced in the target image is obtained from the source image and scaled with the one or more scaling factors.