Segmentation-based Image Processing System for Large Remote Sensing Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation algorithms face challenges such as high computational complexity, sensitivity to merging sequence, over-segmentation, and difficulty in selecting appropriate parameters, especially when dealing with large remote sensing images or multispectral/hyperspectral data, leading to time-consuming trial-and-error processes and unsatisfactory results.
Innovation Solution
The method employs a selectable scale segmentation approach that applies image smoothing, gradient computation, and watershed transformation, using cumulative relative histograms to determine gradient scale levels for suppressing noise and enhancing feature boundaries, allowing for efficient segmentation of both single-band and multi-band images, including remote sensing data, with a tiling scheme for large images and a preview interface for parameter optimization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If traditional region merging algorithms are used for image segmentation, then segmentation can be achieved, but computational complexity becomes very high (O(n log2 n) or worse)
Solution Approach 1:
The patent changes the computational parameters by using a linear-time region merging algorithm that operates in O(n) time complexity instead of the traditional O(n log2 n). This is achieved by modifying the merging criteria and data structures to enable direct linear scanning and merging operations, thereby maintaining segmentation accuracy while dramatically improving processing speed for large remote sensing images.
Solution Approach 2:
The patent divides the image processing into distinct phases: initial region identification, iterative region merging based on similarity criteria, and final segmentation output. This segmented approach allows each phase to be optimized independently, with the merging phase using efficient linear-time operations to reduce overall computational complexity while preserving segmentation quality.
2Productivity
If simple gray level thresholding is used for segmentation, then processing is fast, but segmentation quality deteriorates in uncontrolled image acquisition environments
Solution Approach 1:
The patent implements dynamic region merging where the merging criteria and similarity thresholds are adaptively adjusted during the processing of different image regions. This allows the algorithm to maintain high processing speed through efficient linear operations while dynamically adapting to varying image conditions (lighting, texture, noise levels) to preserve segmentation accuracy in uncontrolled environments.
Solution Approach 2:
The patent applies different merging criteria and similarity measures to different regions of the image based on local characteristics. Each region is evaluated and merged according to its specific properties (homogeneity, edge strength, texture), allowing simple fast operations in uniform regions while applying more sophisticated criteria only where needed, thus maintaining both speed and accuracy.
3Manufacturing precision
If clustering-based segmentation is used, then segmentation can be performed, but computational complexity increases and strict assumptions about cluster shape are required
Solution Approach 1:
The patent extracts and removes the complex iterative optimization steps and strict Gaussian cluster shape assumptions from the segmentation process. Instead, it uses a simpler linear-time region merging approach that operates directly on image regions without requiring complex clustering algorithms, thereby reducing computational complexity and algorithmic assumptions while maintaining segmentation effectiveness.
4Reliability
If watershed transformation is used for gradient-based segmentation, then connected edges are guaranteed, but over-segmentation occurs
Solution Approach 1:
The patent combines the advantages of gradient-based edge detection with region merging by first identifying regions based on gradient information and then iteratively merging adjacent regions based on similarity criteria. This merging step consolidates over-segmented regions while preserving the connected edge structure initially established by the gradient analysis, thus maintaining edge connectivity while correcting over-segmentation errors.
Data Source
AI summary
A digital image can be processed by an image processing method that calculates a gradient map for the digital image, calculates a density function for the gradient map, calculates a modified gradient map using the gradient map, the density function and the selected scale level, and segments the modified gradient map. Prior to segmenting the modified gradient map, a sub-image of the digital image can be segmented at the selected scale level to determine if the selected scale level will give the desired segmentation.