CNN Document Parsing for Style-Independent Key-Value Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Manually extracting structured data from unstructured documents is expensive, time-consuming, and error-prone, and existing systems lack the ability to accurately identify key-value pairs in documents independently of their specific style, structure, or content.
Innovation Solution
A parsing system utilizing a convolutional neural network to automatically convert unstructured documents into structured key-value pairs by identifying bounding boxes around key-value pairs using machine learning and optical character recognition, with a filtering engine to validate the pairs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual extraction methods are used to obtain structured data from unstructured documents, then data accuracy can be maintained through human judgment, but the process becomes expensive and time-consuming
Solution Approach 1:
The patent replaces manual mechanical extraction processes with an automated system combining CNN-based document analysis and OCR technology. The system automatically identifies document structures, locates key-value pairs, and extracts data without human intervention, thereby reducing time loss while maintaining accuracy through machine learning models trained on document patterns.
Solution Approach 2:
The system enables documents to be processed autonomously by training the CNN model on document structures and patterns. The model learns to automatically identify relevant regions, extract key-value pairs, and structure data without requiring manual guidance for each document, making the extraction process self-service and highly efficient.
2Adaptability or versatility
If conventional OCR systems are used to convert documents to text, then text recognition can be performed, but the systems lack the ability to accurately identify key-value pairs independently of document style, structure, or content
Solution Approach 1:
The patent applies local quality by training the CNN model to recognize different document regions and structures specific to various document types (invoices, receipts, forms). The model adapts to local characteristics of different document styles and formats while maintaining the ability to accurately identify key-value pairs within each specific context, rather than using a uniform approach for all documents.
3Measurement precision
If specialized extraction systems are developed for specific document types, then accuracy for those document types can be improved, but the system complexity increases and generalization to other document types becomes difficult
Solution Approach 1:
The patent creates a universal extraction system where a single CNN model can process multiple document types (invoices, receipts, forms, etc.) by learning common structural patterns and key-value pair relationships. The system maintains high accuracy across different document types without requiring separate specialized systems, thereby reducing overall system complexity while preserving extraction precision through the model's ability to generalize from training data.
Data Source
Figure 1
Figure 2A~2B
Figure 2C~2D
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for converting unstructured documents to structured key-value pairs. In one aspect, a method comprises: providing an image of a document to a detection model, wherein: the detection model is configured to process the image to generate an output that defines one or more bounding boxes generated for the image; and each bounding box generated for the image is predicted to enclose a key-value pair comprising key textual data and value textual data, wherein the key textual data defines a label that characterizes the value textual data; and for each of the one or more bounding boxes generated for the image: identifying textual data enclosed by the bounding box using an optical character recognition technique; and determining whether the textual data enclosed by the bounding box defines a key-value pair.