Geometric Text Block Extraction for Unstructured Document Reading

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing image reading systems struggle with automated data extraction from unstructured form-like documents, requiring manual intervention and posing data security issues due to the lack of a generic method applicable to various form templates.

Innovation Solution

An image reading system utilizing a machine learning kernel to generate bounding boxes and search paths, coupled with a descriptive linguistics engine, automatically extracts data from unstructured documents by generating a graph representation and identifying target textual block pairs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If manual processes are used for data extraction from unstructured forms, then data security issues are avoided, but time consumption and tediousness increase significantly

Engineering Contradiction:

Improvedata securityVSAvoidtime consumption

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The system performs automated data extraction using OCR and graph representation algorithms, enabling the document processing system to extract data from unstructured forms autonomously without human intervention, thus improving productivity while maintaining security through automated processing

Inventive Principle:

Principle #25Self-service

2Adaptability or versatility

If generic methods are used for data extraction from various form templates, then adaptability improves, but manufacturing precision of extraction accuracy deteriorates

Engineering Contradiction:

Improveapplicability to various form templatesVSAvoidextraction accuracy

Core Design Contradiction:

Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The system segments the document into textual blocks and generates a graph representation where nodes represent text blocks and edges represent spatial relationships. This segmentation allows the generic method to adapt to various form templates while maintaining extraction accuracy through structured graph analysis

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system changes parameters such as search path directions (horizontal, vertical, diagonal) and bounding box coordinates dynamically based on the document structure. This allows the generic extraction method to adapt to different form templates while maintaining high extraction accuracy through parameter optimization

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If complex graph representation and search path algorithms are used, then extraction accuracy improves, but device complexity increases

Engineering Contradiction:

Improveextraction accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system introduces a graph representation as an intermediary data structure between the raw document image and the extracted data. This graph structure with nodes and edges serves as a mediator that simplifies the extraction process by organizing spatial relationships, thereby improving accuracy without proportionally increasing system complexity

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260080707A1Image reading systems, methods and storage medium for performing geometric extraction

Publication Date: 2026.03.19 US BANK NATIONAL ASSOCIATION

AI summary

Geometric extraction is performed on an unstructured document by recognizing textual blocks on at least a portion of a page of the unstructured document, generating bounding boxes that surround and correspond to the textual blocks, determining search paths having coordinates of two endpoints and connecting at least two bounding boxes, and generating a graph representation of the at least a portion of the page, the graph representation including the plurality of textual blocks, the coordinates of the vertices of each bounding box and the coordinates of the two endpoints of each search path.