Table Recognition System for PDF Structure Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Digital representations of tables in unstructured formats like PDFs lack explicit structure information, limiting their use in computer systems and preventing the extraction of valuable data for automated applications.

Innovation Solution

A computer-implemented method that renders document pages as images, detects tables and their objects using machine learning models, transforms these objects into a structured representation, and exports the data into a desired format, utilizing convolutional neural networks and transformer models for robust table detection and interpretation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If tables are stored in unstructured formats like PDF or images, then flexibility and ease of creation are improved, but structure information is lost and automated processing becomes difficult

Engineering Contradiction:

Improveease of creationVSAvoidstructure information

Core Design Contradiction:

Ease of manufactureVSLoss of information

Solution Approach 1:

The patent introduces an intermediary processing system that converts unstructured table images into structured data representations. The system uses object detection models to identify table components and a post-processing module to reconstruct the structured format, serving as a mediator between unstructured input and structured output requirements

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent replaces manual table structure analysis with automated machine learning models. Convolutional neural networks and transformer models automatically detect table objects, columns, rows, and cells, substituting manual mechanical processing with intelligent automated systems

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Ease of operation

If tables are stored in unstructured formats, then ease of transmission is improved, but data extraction and automated application use become limited

Engineering Contradiction:

Improveease of transmissionVSAvoiddata extraction efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The patent performs preliminary structure recognition and data extraction by converting table images to structured formats in advance. This preliminary processing enables downstream applications to directly use the extracted data without manual intervention, improving overall system productivity

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system enables automated self-service data extraction where the table image automatically undergoes structure recognition, object detection, and data extraction without human intervention. The extracted structured data can then be directly utilized by automated downstream applications

Inventive Principle:

Principle #25Self-service

3Shape

If PDF format is used for tables, then presentation quality is improved, but no markup information exists for structure detection

Engineering Contradiction:

Improvepresentation qualityVSAvoidstructure detection complexity

Core Design Contradiction:

ShapeVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary object detection model that bridges the gap between PDF presentation format and structure detection requirements. The model acts as a mediator that interprets visual table structures without requiring native markup information

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the detection parameters from looking for markup tags (which don't exist in PDF) to detecting visual objects and their spatial relationships. The system transforms the detection task from markup-based to image-based object recognition, adapting to the PDF format's characteristics

Inventive Principle:

Principle #35Parameter changes

4Measurement precision

If joint modeling of table structure recognition and interpretation is used, then accuracy is improved, but computational complexity increases

Engineering Contradiction:

Improvedetection accuracyVSAvoidmodeling complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges table structure recognition and interpretation into a single unified model. By combining these functions, the system achieves better accuracy through joint learning while managing complexity through integrated architecture design rather than separate processing stages

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS12260662B2Inferring structure information from table images

Publication Date: 2025.03.25 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

A computer implemented method includes rendering a document page as an image; detecting tables, columns, and other associated table objects within the image via one or more table recognition models that model objects in the image as overlapping bounding boxes; transforming the set of objects into a structured representation of the table; extracting data from the objects into the structured representation; and exporting the table into the desired output format.