Auto-Encoder Verification for Industrial OCR Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Optical character recognition (OCR) systems in industrial environments face challenges with false positives and 'confusable characters', particularly due to limited training data, requiring significant expertise and time for parameter adjustment and error correction.
Innovation Solution
The implementation of an auto-encoder processor that generates a reconstruction of input image data to validate character classification results, using a deep learning neural network to determine reconstruction errors and distinguish true positives from false positives, thereby improving the accuracy and robustness of OCR systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional OCR systems are used with limited training data, then the system can be deployed quickly, but false positives and confusable characters occur frequently
Solution Approach 1:
The patent applies preliminary action by pre-training the OCR system with a large dataset of confusable characters and their contextual patterns before deployment. This advance preparation enables the system to automatically distinguish between similar characters without requiring time-consuming parameter adjustment during actual use, thereby improving reliability while minimizing time loss.
Solution Approach 2:
The patent implements feedback mechanisms where the system continuously learns from verification results and adjusts its classification thresholds. By analyzing false positive patterns and incorporating contextual information from surrounding text, the system refines its accuracy over time without requiring manual retraining, thus improving reliability without proportional time investment.
2Reliability
If OCR parameters are adjusted to reduce false positives, then accuracy improves, but system complexity and expertise requirements increase
Solution Approach 1:
The patent applies self-service by enabling the OCR system to automatically adjust its own parameters through machine learning algorithms. The system autonomously identifies confusable character patterns and adjusts classification thresholds based on contextual analysis, eliminating the need for manual parameter tuning by experts and reducing system complexity from the user's perspective.
Solution Approach 2:
The patent transforms the complex task of manual parameter adjustment into automatic parameter optimization through adaptive algorithms. By changing from static, manually-tuned parameters to dynamic, self-adjusting parameters based on contextual patterns, the system achieves high accuracy without increasing operational complexity or requiring expert knowledge.
3Reliability
If extensive training data is collected to improve OCR robustness, then false positives are reduced, but data collection and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-collecting and pre-processing a comprehensive dataset of confusable characters and their contextual patterns before system deployment. This advance data preparation creates a robust foundation that enables the system to handle diverse characters without requiring additional data collection during operation, thereby improving robustness while minimizing time loss.
Solution Approach 2:
The patent uses copying by creating synthetic training examples through transformation and augmentation of existing data. By generating varied versions of confusable character pairs through geometric transformations, noise addition, and contextual variations, the system achieves robustness with less actual data collection time while maintaining high reliability.
Data Source
AI summary
An auto-encoder is configured to verify character detection and/or classification results generated by an automated optical character recognition system. The auto-encoder may be trained to reconstruct visual representations of the detected character, and a determination of whether the character detection result comprises a true positive or false positive may be based on a reconstruction error between the image data in which the character was detected and a reconstructed image generated by the auto-encoder.