Polygon Extraction from Distorted Document Images Using Vertex Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods fail to accurately extract a portion of an invoice or similar documents from images when they are imaged at an angle, leading to issues like breakage or wrinkles, which affect the shape and make it difficult to appropriately extract the desired rectangular or polygonal regions.
Innovation Solution
An image processing system that includes an input unit, detection unit, convex hull generating unit, vertex reducing unit, and extraction unit to detect and extract a polygonal portion from an image, using a vertex reducing process to minimize the area of the extracted region and ensure it encompasses the entire target, even with distortions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a conventional histogram-based method is used to extract rectangular portions, then the extraction process is simple, but it fails to accurately extract portions when the invoice is imaged at an angle or has breakage and wrinkles
Solution Approach 1:
The patent segments the extraction process into multiple distinct steps: detecting contour points of the invoice, generating a convex hull from these points, identifying vertices of the convex hull, and reducing vertices to obtain the final extraction region. This segmentation allows each step to be optimized independently, improving overall extraction accuracy while managing complexity through modular processing
Solution Approach 2:
The patent performs preliminary actions by first detecting all contour points of the invoice before extraction, and generating a convex hull that encompasses the entire invoice area. This preliminary convex hull generation ensures that even if the invoice has breakage or wrinkles, the initial extraction region covers the complete target area, allowing subsequent vertex reduction to refine the shape without losing target content
2Adaptability or versatility
If the extraction method assumes rectangular shapes only, then the processing is straightforward, but it cannot handle non-rectangular quadrilaterals caused by angled imaging
Solution Approach 1:
The patent implements a dynamic vertex reduction process that adaptively adjusts the extraction polygon based on the detected contour points and convex hull vertices. Instead of assuming a fixed rectangular shape, the system dynamically determines the number of vertices and their positions by reducing vertices of the convex hull based on area minimization criteria, allowing the extraction region to adapt to various shapes including non-rectangular quadrilaterals while maintaining precision
Solution Approach 2:
The patent changes the parameter of shape constraints by transitioning from fixed rectangular assumptions to variable polygon shapes with adaptive vertex counts. The vertex reduction process modifies the polygon parameters dynamically based on the actual invoice contour, allowing the extraction region to conform to the true shape of the invoice regardless of imaging angle or physical condition
3Manufacturing precision
If vertices are reduced to minimize the extracted area, then the extraction becomes more precise, but the process becomes more complex
Solution Approach 1:
The patent replaces complex mechanical or manual extraction methods with an automated computational vertex reduction algorithm. The algorithm systematically identifies vertices of the convex hull, calculates area contributions of each vertex, and removes vertices that minimize the total area while maintaining coverage of the invoice content. This substitution of automated computation for manual or mechanical processes achieves high precision without proportionally increasing operational complexity
Data Source
AI summary
A portion of a polygon is appropriately extracted from an image. An image processing system 10 including: an input unit 11 that inputs an image including an extraction target; a detection unit 12 that detects a portion of the extraction target from an input image; a convex hull generating unit 13 that generates a convex hull, which is a polygonal, of the detected portion; a vertex reducing unit 14 that acquires a polygon having a preset number of vertices from the convex hull by repeating a vertex reducing process of eliminating a vertex according to each interior angle of the polygon; and an extraction unit 15 that extracts a portion of the acquired polygon described above from the input image, in which the vertex reducing process is a process of determining a new line based on an area of a region surrounded by the new line passing through a vertex to be eliminated, two sides generating the vertex in the polygon, and two lines acquired by extending two adjacent sides of the vertex in the polygon and generating a polygon after elimination of the vertex using the determined new line and the two extended lines.