Document Image Classification and OCR for Expense Item Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems require extensive machine learning and user-specific data preparation for journalizing expense items, and character recognition errors occur frequently due to unclear company names or special fonts in document images, complicating automated data entry tasks.
Innovation Solution
An image processing apparatus and method that includes a character recognizing unit for OCR and a first identifying unit to classify document images based on predefined conditions, using dictionaries and pattern extraction to identify company classification types and extract item values, even when specific names are unclear.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If machine learning is performed in advance to extract journalizing elements from document images, then the automation of data entry tasks is improved, but the requirement for extensive learning data and user-specific data preparation increases system complexity
Solution Approach 1:
The system performs preliminary classification of document images into categories (invoices, receipts, bank statements, etc.) using predefined conditions and patterns before extracting specific journalizing elements. This preliminary action organizes the processing flow and enables the use of specialized extraction rules for each document type, reducing the need for universal machine learning models trained on extensive diverse data.
Solution Approach 2:
The character recognition process is segmented into multiple passes: first recognizing high-frequency standard characters, then recognizing remaining characters using contextual patterns and dictionaries. This segmentation allows the system to handle uncertain characters by leveraging surrounding context and predefined patterns, reducing recognition errors without requiring extensive training data for all possible character variations.
2Loss of information
If character recognition processing is performed on document images with unclear company names or special fonts, then the extraction of company information is attempted, but recognition errors increase
Solution Approach 1:
The system introduces an intermediary classification step that categorizes document images before extracting company information. By classifying the document type first (invoice, receipt, bank statement, etc.), the system can apply specialized extraction patterns and dictionaries appropriate for each type, improving the reliability of company information extraction even when company names are unclear or use special fonts.
Solution Approach 2:
The character recognition process dynamically adjusts recognition parameters based on document classification. For example, after classifying a document as an invoice, the system can apply invoice-specific patterns and dictionaries for company name extraction, changing the recognition approach to match the document type and thereby improving accuracy for that specific category.
3Adaptability or versatility
If rules for journalizing expense items are customized for each user, then the adaptability to different business requirements is improved, but the requirement for user-specific learning data increases time consumption
Solution Approach 1:
The system implements a universal classification framework that handles multiple document types (invoices, receipts, bank statements, etc.) with a single unified approach. Each document type has predefined extraction patterns and dictionaries that work across different users' requirements, providing adaptability without requiring each user to prepare custom learning data. Users can configure their preferred journalizing rules, but the underlying extraction engine remains universal.
Data Source
AI summary
The present application is to obtain a character recognition result by performing character recognition processing on a document image and identify a classification type of the document image based on a character string included in the character recognition result and a predefined condition. The condition for identifying classification types that are hints for expense items is defined in advance.