Document Border Detection Using Line Weight Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies face challenges in accurately identifying lines in scanned images, particularly in distinguishing meaningful lines from stray marks, and efficiently storing and using this information for image correction and cropping, especially when dealing with distorted rectangular documents.
Innovation Solution
The method involves detecting lines in an image, classifying them based on direction, determining corner points, and applying corrections to transform a non-rectangular image into a rectangular format, using parameter-accumulator arrays and line-weight calculations to identify and weight lines for effective edge detection and image processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If line detection is performed on scanned images to identify document edges, then document border detection accuracy is improved, but storage requirements for line information increase significantly
Solution Approach 1:
The patent extracts only the essential line information needed for document border detection, specifically storing line parameters (slope, intercept, weight) rather than complete line data. This selective extraction reduces storage requirements while maintaining detection accuracy by focusing on the most relevant features for identifying document edges.
Solution Approach 2:
Instead of storing all detected lines and then filtering for border lines, the patent inverts the approach by using weight calculations during line detection to prioritize and identify border lines directly. Lines with higher weights (indicating stronger edge characteristics) are automatically selected as potential document borders, reducing the need to store and process all line data.
2Loss of information
If all detected lines are stored for subsequent processing, then line information completeness is improved, but processing efficiency decreases due to large data volumes
Solution Approach 1:
The patent applies local quality by assigning different weights to different lines based on their characteristics (slope, position, continuity). Instead of treating all lines uniformly, lines with higher weights (indicating better edge properties) are prioritized for storage and processing. This selective approach maintains essential line information while reducing overall data volume, thereby improving processing efficiency without significant loss of critical information.
Solution Approach 2:
The patent changes the parameter representation of lines by storing only key parameters (weight, slope, intercept) rather than complete line data. This parameter transformation reduces the amount of data that needs to be stored and processed while preserving the essential characteristics needed for document border detection and image correction operations.
3Measurement precision
If automated line classification is implemented to distinguish meaningful lines from stray marks, then line identification accuracy is improved, but algorithm complexity increases
Solution Approach 1:
The patent implements self-service by allowing lines to automatically classify themselves through weight calculation based on their inherent characteristics. Lines with strong edge properties naturally accumulate higher weights during the detection process, automatically distinguishing them from stray marks without requiring complex external classification algorithms. This approach maintains high identification accuracy while keeping the algorithm relatively simple.
Solution Approach 2:
The patent uses feedback mechanisms where line detection results are fed back into the weighting process. Lines that consistently appear across multiple detection passes or exhibit consistent characteristics receive higher weights, allowing the system to automatically distinguish meaningful lines from noise. This iterative feedback approach improves identification accuracy without requiring overly complex algorithms.
Data Source
AI summary
Methods and apparatus for: identifying perimeter lines in an image, e.g., document border lines forming a non-rectangular quadrilateral corresponding to a scanned document, determining document corner points, and correcting the input image based on determined corner points are described. Lines are identified in the input image and a line weight is generated for each identified line. The identified lines are classified into two different groups based on direction, e.g., predominately vertical or predominately horizontal. For each group at least two, and sometimes more, candidate perimeter lines are determined to be kept, based on line weight and line length, and other lines in the group are eliminated. Two best candidate perimeter lines are identified from each group, e.g., based on distance between the remaining lines in the group. Four document corner points are determined based on the remaining lines in the two groups.