Document Corner Detection Using Bounding Rectangle Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in accurately detecting the corners of documents with rounded corners, leading to inefficiencies in perspective correction and increased processing errors, particularly when capturing images for verification purposes.
Innovation Solution
A method involving a user device that captures images using a camera, detects edges and lines, filters candidate lines and points, and identifies the four closest corner points to a bounding rectangle to perform perspective correction, which includes tilt and rotation adjustments, ensuring accurate corner detection and image representation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing image processing systems are used to detect document corners, then the process can be completed, but the corner detection accuracy is insufficient especially for documents with rounded corners
Solution Approach 1:
The patent segments the corner detection process into multiple stages: first detecting edges, then detecting lines from edges, identifying intersection points, filtering candidate points, and finally selecting the four closest points to bounding rectangle corners. This segmentation allows each stage to optimize for specific aspects of accuracy without compromising overall reliability.
Solution Approach 2:
The patent performs preliminary actions by detecting edges and lines before identifying corner candidates. By pre-processing the image to extract edge information and line segments, the system establishes a foundation that guides subsequent corner detection, ensuring accuracy even for rounded corners before the final selection is made.
2Measurement precision
If comprehensive line detection is performed to improve corner detection accuracy, then more corner candidates are identified, but processing time increases
Solution Approach 1:
The patent extracts only the necessary information needed for corner detection by filtering lines based on their intersection with detected edges. Instead of processing all possible lines in the image, the system extracts only those that intersect with edge segments, significantly reducing the number of candidates while maintaining detection accuracy.
Solution Approach 2:
The patent applies partial action by detecting lines only in regions where edges are present and using only the intersection points that fall within the bounding rectangle. This selective approach processes only the necessary portions of the image data, reducing overall processing time while maintaining sufficient accuracy.
3Reliability
If multiple lines and intersection points are detected to ensure accurate corner identification, then corner detection becomes more reliable, but the complexity of the processing system increases
Solution Approach 1:
The patent implements feedback mechanisms at multiple stages: edge detection feeds into line detection, line intersections feed into candidate point identification, and the bounding rectangle geometry feeds back into filtering the final four corner points. This feedback loop ensures reliability by continuously validating results against geometric constraints.
Solution Approach 2:
The patent introduces intermediary structures such as the bounding rectangle and filtered candidate point sets that mediate between the complex line detection process and the final corner identification. These intermediaries simplify the overall process by providing clear criteria for selection without requiring complex direct calculations.
Data Source
AI summary
In some implementations, a device may obtain an image of an object using a camera of the device and identify at least two corner points from a set of corner candidate points closest, respectively, to at least one corner of a bounding shape used to generate the set of corner candidate points. The device may validate the at least two corner points by determining that at least one edge length between the at least two corner points is within a threshold size of a corresponding edge of the bounding shape. Furthermore, the device may automatically trigger image capture based on satisfying predetermined conditions related to the identified at least two corner points.