Region Extraction Apparatus Using Graph Cut Contouring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional region extraction methods in image processing struggle to accurately extract specific regions, such as water droplets on a leaf, due to noise detection and false contour identification, especially when threshold values are lowered to increase detection numbers.
Innovation Solution
A region extraction apparatus that performs thresholding on pixel intensity values to identify high and low intensity regions, determines their combination based on arrangement positions, and extracts contours within an outer frame using a graph cut algorithm after intensity normalization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If threshold value is lowered to increase the number of detections, then the quantity of detected regions increases, but the measurement precision decreases due to noise detection and false contour identification
Solution Approach 1:
The patent segments the image processing into distinct stages: initial thresholding to capture multiple candidate regions, followed by contour extraction and region-of-interest determination. This multi-stage segmentation allows the system to initially cast a wide net for detections, then progressively refine and filter results to eliminate noise and false positives, thereby resolving the contradiction between quantity and precision.
Solution Approach 2:
The patent performs preliminary contour extraction and region-of-interest determination before final region identification. By extracting contours first and determining which regions contain actual objects (versus noise), the system prepares the data structure in advance to facilitate accurate filtering. This preliminary action enables the system to maintain high detection quantities while ensuring precision in the final output.
2Ease of manufacture
If conventional contour extraction methods are used, then the device complexity is low, but the manufacturing precision decreases due to inability to accurately extract regions like water droplets on leaves
Solution Approach 1:
The patent performs preliminary contour extraction using graph cuts or similar methods to identify potential region boundaries before determining regions of interest. This preliminary contour extraction creates a structured framework that subsequent processing can refine, maintaining relative simplicity while improving accuracy for difficult cases like water droplets on leaves.
Solution Approach 2:
The patent introduces an intermediary region-of-interest determination step between initial thresholding and final region extraction. This intermediary process uses the extracted contours and image data to identify which areas contain actual objects versus noise, acting as a mediator that bridges simple extraction methods with accurate region identification, thereby improving precision without proportionally increasing complexity.
3Device complexity
If only a few feature points are detected in images like water droplets on leaves, then the device complexity remains low, but the productivity decreases due to insufficient number of similar regions obtained
Solution Approach 1:
The patent segments the detection process into multiple passes: initial thresholding to capture all potential regions, contour extraction to define boundaries, and region-of-interest determination to identify valid objects. This segmentation allows the system to progressively increase the number of detected regions without requiring complex algorithms at each stage, thereby improving productivity while controlling complexity.
Solution Approach 2:
The patent performs preliminary contour extraction and region-of-interest determination to prepare comprehensive candidate data before final similar region search. By having contours and ROI information ready in advance, the system can efficiently perform similarity searches across many regions without increasing the complexity of the core detection algorithms, thus improving productivity.
Data Source
AI summary
A region extraction apparatus includes: a region extraction unit that extracts, from an image data of a process target, a high intensity region having higher intensity than a neighboring region thereof and a low intensity region having lower intensity than a neighboring region thereof; a combination identification unit that identifies a combination of a high intensity region and a low intensity region corresponding to an extraction target region to be extracted based on arrangement positions of the high intensity region and the low intensity region extracted by the region extraction unit; an outer frame determination unit that determines an outer frame corresponding to the combination of the high intensity region and the low intensity region; and a contour extraction unit that extracts a contour of the extraction target region based on image data within the outer frame determined by the outer frame determination unit.