OCR Pre-Processing Neural Network Rotation Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical character recognition (OCR) processes face errors due to information loss or distortion when dealing with documents that are rotated or scaled differently from the expected parameters, leading to time-intensive and computationally costly sequential pre-processing steps.
Innovation Solution
A machine learning system, utilizing a convolutional neural network, identifies and corrects document image parameters such as rotation and scaling to standardize the image before OCR processing, using a shared computational graph for faster and more efficient character recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sequential pre-processing steps (de-skewing, de-speckling, binarization, line removal, layout analysis) are used for OCR, then character recognition can be performed, but information loss or distortion occurs especially when documents are rotated or scaled differently from expected parameters
Solution Approach 1:
The system performs preliminary action by detecting document rotation and scale parameters before the main OCR processing. The convolutional neural network analyzes the document image to identify orientation and scaling characteristics, then applies corrective transformations to align the document with expected OCR parameters, preventing information loss before recognition begins
Solution Approach 2:
The system changes parameters by dynamically adjusting rotation and scale transformations based on detected document characteristics. The neural network determines optimal transformation parameters (rotation angles, scale factors) to correct document orientation and sizing, converting the document to match expected OCR input parameters while preserving character integrity
2Reliability
If sequential pre-processing steps are used for OCR, then character recognition can be performed, but the process becomes time-intensive with significant computational costs
Solution Approach 1:
The system merges previously separate sequential processing steps into a unified end-to-end deep learning model. The convolutional neural network integrates rotation detection, scale detection, and OCR recognition into a single integrated architecture that processes the document in one continuous operation, eliminating the time delays associated with sequential step-by-step processing
Solution Approach 2:
The system achieves universality by designing a multi-functional neural network that simultaneously performs multiple tasks: detecting rotation, detecting scale, and performing character recognition. This single unified model replaces multiple specialized pre-processing modules, reducing computational overhead and processing time while maintaining comprehensive functionality
3Reliability
If sequential pre-processing steps are used for OCR, then character recognition can be performed, but the device complexity increases with multiple processing modules
Solution Approach 1:
The system merges multiple complex pre-processing modules into a single convolutional neural network architecture. Instead of implementing separate de-skewing, de-speckling, binarization, line removal, and layout analysis modules, the unified neural network performs all these functions simultaneously through integrated layers, significantly reducing system complexity while maintaining processing capability
Data Source
AI summary
Disclosed herein are system, method, and computer program product embodiments for optical character recognition (OCR) pre-processing using machine learning. In an embodiment, a neural network may be trained to identify a standardized document rotation and scale expected by an OCR service performing character recognition. The neural network may then analyze a received document image to identify a corresponding rotation and scale of the document image relative to the expected standardized values. In response to this identification, the document image may be modified in the inverse to standardize the rotation and scale of the document image to match the format expected by the OCR service. In some embodiments, a neural network may perform the standardization as well as the character recognition using a shared computation graph.