Neural Network Table Boundary Identification in Spreadsheets

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies face challenges in automatically identifying table boundaries in spreadsheets due to the variety of visual information and data present, leading to limited success with rule-based approaches.

Innovation Solution

The use of a machine learning-based method that involves receiving a spreadsheet, identifying classes for each cell using a neural network (specifically, corner and not-a-corner classes), and inducing table boundaries based on these classifications.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If rule-based approaches are used to identify table boundaries, then the method is simple and interpretable, but the identification accuracy is limited due to the variety of visual information and data in spreadsheets

Engineering Contradiction:

Improvetable boundary identification accuracyVSAvoidmethod complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces rule-based mechanical classification methods with a neural network-based machine learning system. The neural network automatically learns complex patterns and features from spreadsheet images, substituting the manual rule-based approach with an adaptive computational model that achieves superior identification accuracy while handling the variety of visual information in spreadsheets

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent transforms the table boundary identification problem into a corner classification problem with specific classes (top-left, top-right, bottom-left, bottom-right corners). By changing the parameter space from general boundary detection to specific corner point classification, the system achieves more precise and reliable table boundary identification

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If a neural network-based machine learning approach is used, then identification accuracy improves, but computational complexity and resource requirements increase

Engineering Contradiction:

Improvetable boundary identification accuracyVSAvoidcomputational resource consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments the spreadsheet image into individual cell regions and processes them independently through the neural network. This segmentation approach allows parallel processing of multiple cells, reducing the overall computational burden while maintaining high identification accuracy for each cell's corner classification

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent extracts and focuses on corner cells as the key discriminative features for table boundary identification. By taking out only the corner cells for detailed classification while treating other cells more simply, the system reduces computational resource consumption while maintaining high accuracy in identifying table boundaries

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20250068837A1Systems, methods, and computer-readable media for improved table identification using a neural network

Publication Date: 2025.02.27 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Systems, methods, and computer-readable storage devices are disclosed for improved table identification in a spreadsheet. One method including: receiving a spreadsheet including at least one table; identifying, using machine learning, one or more classes of a plurality of classes for each cell of the received spreadsheet, wherein the plurality of classes include corners and not-a-corner; and inducing at least one table in the received spreadsheet based on the one or more identified classes for each cell of the received spreadsheet.