Rotating Tables in Images for OCR Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical character recognition (OCR) algorithms face challenges when dealing with rotated tables in images, leading to incorrect results due to the rotation of text characters with respect to the horizontal or vertical axis.
Innovation Solution
A method and system for image processing that identifies lines in a table, calculates bin values for angle bins based on confidence values, and rotates the image to align the table closer to horizontal or vertical, improving OCR accuracy by using a buffer, line extractor, angle bin engine, and table engine to correct the rotation and enhance OCR performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If OCR algorithms are applied to rotated tables in images, then text character identification can be performed, but the rotation causes incorrect recognition results
Solution Approach 1:
The patent applies preliminary action by detecting the rotation angle of the table and rotating the image to correct the orientation before performing OCR recognition. This preprocessing step ensures that text characters are properly aligned with the horizontal axis, eliminating the recognition errors caused by rotation while maintaining the ability to process rotated tables
2Reliability
If the image is rotated to align the table horizontally, then OCR accuracy improves, but additional image processing steps are required
Solution Approach 1:
The patent applies self-service by implementing an automated system that autonomously detects table rotation angles, calculates the necessary correction, rotates the image accordingly, and performs OCR. The system serves itself by integrating all these functions into a unified processing pipeline, reducing the need for manual intervention while improving recognition accuracy
Data Source
AI summary
A method for image processing, including: obtaining an image including a table; identifying a first line corresponding to the table in the image, a first confidence value for the first line, and a first angle for the first line; initiating a plurality of angle bins corresponding to multiple angles; calculating, based on the first confidence value, a first plurality of bin values for a first subset of the plurality of angle bins within a window surrounding the first angle; adding the first plurality of bin values to the first subset of the plurality of angle bins; identifying an angle bin of the plurality of angle bins having a maximum bin value; and rotating the image based on the angle bin having the maximum bin value.