Domain-Specific OCR Training Images for Specialized Text Accuracy

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Optical Character Recognition (OCR) machine learning models struggle with low accuracy for domain-specific text, particularly handwritten text on ruled notepad paper, due to insufficient training data and varied image types.

Innovation Solution

Generate domain-specific images using web scraping techniques to identify relevant words and phrases, applying various fonts, backgrounds, and filters to create training data tailored to the domain, improving model accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If general-purpose training data is used, then the OCR model can recognize common text, but it fails to accurately recognize domain-specific words

Engineering Contradiction:

ImproveOCR accuracy for domain-specific textVSAvoidAbility to handle multiple domains

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent segments the training data into domain-specific subsets by identifying and separating domain-relevant words from general text corpora. This is achieved through web scraping targeted at specific domains and filtering words based on domain relevance, allowing the model to be trained on specialized vocabulary without compromising general text recognition capabilities.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by creating domain-specific training images with customized characteristics tailored to each domain's requirements. This includes using domain-appropriate fonts, backgrounds, and text styles that match the specific domain's visual characteristics, thereby improving recognition accuracy for that particular domain while maintaining versatility across multiple domains through repetition of the process for different domains.

Inventive Principle:

Principle #3Local quality

2Reliability

If domain-specific training data is collected manually, then the model learns domain vocabulary, but the process is time-consuming and resource-intensive

Engineering Contradiction:

ImproveOCR accuracy for domain-specific textVSAvoidTime for data collection and preparation

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-scraping and storing domain-relevant words from web sources before the actual OCR training process. This involves creating a domain vocabulary database in advance, which can be reused across multiple training iterations and domains, significantly reducing the time required for data collection during model training.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system implements self-service by automatically scraping web content, extracting domain-relevant words, and generating training images without requiring manual data collection. The automated pipeline includes web scraping, word frequency analysis, and image generation that operates independently, eliminating the need for manual intervention in data preparation while maintaining high domain-specific accuracy.

Inventive Principle:

Principle #25Self-service

3Reliability

If the training data does not include domain-specific visual characteristics, then the model lacks domain knowledge, but creating realistic domain-specific images is complex

Engineering Contradiction:

ImproveRecognition of domain-specific text characteristicsVSAvoidComplexity of image generation process

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by systematically varying image generation parameters such as font type, background color, text size, and style to match domain-specific characteristics. This allows the creation of realistic domain-specific training images through controlled parameter adjustments rather than complex manual image creation, simplifying the process while maintaining authenticity.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent uses copying by generating synthetic training images that replicate the visual characteristics of real domain-specific text. Instead of manually creating or collecting authentic domain images, the system copies the essential visual properties (fonts, backgrounds, layouts) through automated image generation, reducing complexity while preserving domain-specific features needed for accurate recognition.

Inventive Principle:

Principle #26Copying

Data Source

PatentUS12633149B2Generation of domain-specific images for training optical character recognition (OCR) machine learning model

Publication Date: 2026.05.19 MICRO FOCUS LLC

AI summary

Domain-specific images used for training an optical character recognition (OCR) machine learning model are generated as follows. Universal resource locator (URL) addresses of web pages associated with a particular domain are retrieved. Words in the web pages associated with the particular domain are determined. Domain-relevant n-grams of the words are identified for the particular domain. Corresponding domain-specific images of each domain-relevant n gram for the particular domain are generated.