Object-to-Text Conversion Using Graph-Based Positional Relations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately convert schematic views, product drawings, and patent diagrams into text, often leading to variations in positional relations and difficulties in determining overlapping or non-overlapping states of objects.
Innovation Solution
A method and system that utilize a computer device to extract coordinate data, convert positional relations into graph data, and generate text from objects, including a GUI, image processing, feature extraction, and text generation units to accurately describe the positional relations between objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If coordinate data extraction and positional relation conversion methods are used to convert objects into text, then text generation accuracy is improved, but system complexity increases due to multiple processing units and graph data conversion steps
Solution Approach 1:
The system is divided into distinct functional units: image processing unit for coordinate extraction, feature extraction unit for positional relation analysis, graph generation unit for relationship modeling, and text generation unit for output creation. This segmentation allows each unit to specialize in a specific task, improving overall accuracy while making the complexity manageable through modular design.
Solution Approach 2:
Graph data serves as an intermediary representation between the extracted coordinate data and the final text output. The graph generation unit creates structured relationship data that captures positional relations, which then serves as input for the text generation unit. This intermediary step enables accurate representation of complex spatial relationships before text generation.
2Speed
If AI recognition methods are used to identify objects in images, then recognition speed is improved, but recognition accuracy deteriorates for objects or components that AI has never learned
Solution Approach 1:
The system replaces AI-based recognition with a deterministic coordinate extraction approach. Instead of using neural networks to identify and classify objects, the image processing unit directly extracts coordinate data from the image, and the feature extraction unit determines positional relationships based on geometric calculations. This substitution eliminates AI's limitation with unseen objects while maintaining fast processing through algorithmic operations.
3Loss of time
If simple text generation methods are used without graph data conversion, then processing time is reduced, but positional relation accuracy deteriorates due to variations in defining spatial relationships
Solution Approach 1:
The graph generation unit performs preliminary processing by converting coordinate data into structured graph representations that explicitly encode positional relationships before text generation. This pre-processing step organizes spatial information into a standardized format, making subsequent text generation more accurate and efficient. The graph structure captures relationships that would otherwise require complex text processing.
Data Source
AI summary
Text is generated from an object. Text is generated from a first object. The first object includes a second object and a third object. A step of detecting coordinate data of the second object is included. A step of detecting coordinate data of the third object is included. A step of extracting positional relation between the second object and the third object from coordinate data is included. A step of converting the extracted positional relation into graph data is included. A step of generating text about the positional relation between the second object and the third object from graph data is included.