Multi-layer RNN and CNN for Context Sensitive OCR
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Convolutional neural networks (CNN) used for optical character recognition (OCR) often struggle to accurately distinguish between similar characters, leading to inaccuracies that can impact downstream natural language processing and document processing applications.
Innovation Solution
A computing platform trains multiple layers of recursive neural networks (RNN) in conjunction with a CNN, using historical images and contexts at different levels (word, sentence, and document) to enhance OCR accuracy by iteratively refining character identification based on confidence scores, with manual input requested when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a CNN is used to perform OCR, then the system can process images and extract text, but the CNN may identify multiple character candidates and be unable to distinguish which is correct, resulting in inaccuracies
Solution Approach 1:
The patent implements a nested architecture where multiple RNN layers are integrated within the CNN-based OCR system. Each RNN layer processes the output of the previous layer, with the first RNN layer receiving CNN output, the second RNN layer receiving output from the first, and so on. This nested structure allows progressive refinement of character recognition through multiple contextual analysis stages, resolving the contradiction by maintaining high accuracy while improving reliability through iterative validation.
Solution Approach 2:
The system implements feedback mechanisms where each RNN layer analyzes the output of the previous layer and provides refined results back to the processing pipeline. The confidence scores generated at each layer feed into the next layer's decision-making process, allowing the system to iteratively improve character identification accuracy and reliability through multiple passes of contextual analysis.
2Measurement precision
If multiple RNN layers are added to improve OCR accuracy through contextual analysis, then recognition precision improves, but the system complexity increases
Solution Approach 1:
The patent segments the contextual analysis function into multiple specialized RNN layers, each responsible for a specific level of text hierarchy (character level, word level, sentence level). This segmentation allows the complex task of contextual analysis to be divided into manageable, modular components that can be processed sequentially, reducing overall system complexity while maintaining high accuracy through specialized processing at each stage.
Solution Approach 2:
The system adds a temporal dimension to the OCR process by introducing sequential RNN layers that process information over time. Instead of a single static analysis pass, the multi-layer RNN architecture processes contextual information across multiple time steps, allowing the system to capture hierarchical dependencies in text data without significantly increasing spatial complexity of the network architecture.
Data Source
AI summary
Aspects of the disclosure relate to OCR. A computing platform may train, using historical images, a CNN and a RNN to perform OCR/identify characters in context. The computing platform may receive an image of a document, and may input the image into the CNN, which may cause the CNN to output OCR information for the image and a confidence score. Based on identifying that the confidence score exceeds a confidence threshold, the computing platform may store the OCR information to enable subsequent access of a digital version of the document. Based on identifying that the confidence score does not exceed the confidence threshold, the computing platform may: 1) input the OCR information into the first RNN, which may cause the first RNN to output contextual OCR information for the image, and 2) store the contextual OCR information to enable subsequent access of the digital version of the document.