Document Rotation Detection via OCR Confidence Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated document processing systems face challenges in accurately detecting and correcting the orientation of document images, which are often scanned or uploaded in arbitrary rotations, leading to degraded model accuracy and inefficient training.
Innovation Solution
A method is proposed that involves rotating document images by multiple angles, performing optical character recognition analysis, and generating confidence scores to associate rotation classifications. A machine learning model is then trained using a combination of rotated images with known orientations and images with estimated orientations based on OCR analyses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to create training data, then labeling accuracy is improved, but time consumption and labor costs increase significantly
Solution Approach 1:
The system uses automated OCR-based labeling where the document processing system labels itself without human intervention. The OCR engine analyzes document images, determines rotation angles, and generates training labels automatically, making the system self-sufficient and eliminating the need for manual annotation while maintaining high labeling accuracy through automated intelligent processing
Solution Approach 2:
The patent replaces the mechanical manual labeling process with an automated computational system. Instead of humans visually inspecting and annotating documents, the system uses optical character recognition algorithms, image processing, and machine learning models to automatically detect rotation angles and generate training data, substituting mechanical human labor with automated digital processing
2Quantity of substance
If synthetic data is generated for training, then data volume is improved, but representativeness of real-world rotation distributions deteriorates
Solution Approach 1:
The system uses OCR analysis of real document images as feedback to generate training data. By analyzing the actual rotation angles found in real-world documents through OCR processing, the system creates training datasets that reflect true rotation distributions. This feedback loop ensures synthetic or augmented data maintains statistical representativeness of real-world conditions while still providing sufficient data volume for robust model training
3Productivity
If document images are processed at arbitrary orientations, then processing speed is maintained, but model accuracy deteriorates
Solution Approach 1:
The system performs preliminary rotation correction before the main document analysis processing. By detecting the rotation angle of each document image and automatically rotating it to the correct upright orientation in advance, the system ensures that subsequent analysis models receive properly oriented images, thereby maintaining high accuracy without requiring models to handle arbitrary orientations
Data Source
AI summary
Certain aspects of the disclosure provide a method for generating training data and training a machine learning model. The method may include rotating each document image in a first set of document images by a plurality of rotation angles to obtain a first set of rotated document images and associating a rotation classification label to each rotated document image in the first set of rotated document images. The method may further include for each document image in a second set of document images: rotating the respective document image by a plurality of rotation angles, performing an optical character recognition analysis at each rotation angle of the plurality of rotation angles, generating a confidence score based on the optical character recognition analyses, assigning the confidence score to the respective document image, and associating a rotation classification to the respective document image based on the optical character recognition analyses.