Graph Modeling for Image Content Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition technologies struggle to accurately extract categorizable information from images, particularly when the data types lack consistent patterns, such as shipment tracking numbers which vary in format across different couriers.
Innovation Solution
A data extraction system that uses a graph modeling approach to analyze the positions and relationships of data within an image, allowing for the identification of categorizable content even in the absence of consistent patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pattern-based OCR techniques are used to extract text from images, then text identification is achieved, but the ability to categorize information reliably deteriorates when patterns are inconsistent
Solution Approach 1:
The patent segments the image into multiple regions of interest (ROIs) containing different types of data elements. Each ROI is processed independently to extract features and relationships, allowing the system to handle inconsistent patterns across different data types while maintaining accurate extraction through localized analysis.
Solution Approach 2:
The patent introduces a graph neural network as an intermediary between the extracted text data and the categorization task. This intermediary learns relationships and contextual patterns from the graph structure, enabling reliable categorization even when traditional text patterns are inconsistent or ambiguous.
2Adaptability or versatility
If traditional image recognition applications rely on character arrangement patterns, then recognizable data types can be identified, but adaptability to new or varied data formats deteriorates
Solution Approach 1:
The patent employs a graph neural network that dynamically adapts to different data formats and relationships. The model can learn new patterns and relationships from the graph structure without requiring explicit programming for each data type, providing versatility while managing complexity through learned representations rather than rigid rules.
Solution Approach 2:
The patent creates a universal graph-based framework that can handle multiple types of data formats and relationships simultaneously. The same graph neural network architecture can process different data types (addresses, phone numbers, tracking numbers, etc.) by learning their specific patterns from the graph structure, eliminating the need for separate specialized systems for each data type.
3Reliability
If graph modeling approach is used to extract spatial relationships, then categorization accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the image into multiple regions of interest (ROIs) containing different types of data elements. Each ROI is processed independently to extract features and relationships, then combined into a graph structure. This segmentation reduces the complexity of processing the entire image at once while maintaining the benefits of graph-based relationship analysis for improved categorization accuracy.
Data Source
AI summary
Methods and systems are presented for extracting categorizable information from an image using a graph that models data within the image. Upon receiving an image, a data extraction system identifies characters in the image. The data extraction system then generates bounding boxes that enclose adjacent characters that are related to each other in the image. The data extraction system also creates connections between the bounding boxes based on locations of the bounding boxes. A graph is generated based on the bounding boxes and the connections such that the graph can accurately represent the data in the image. The graph is provided to a graph neural network that is configured to analyze the graph and produce an output. The data extraction system may categorize the data in the image based on the output.