Document Image Extraction Using Segmented Model and Rule-Based Logic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for extracting character strings corresponding to company names of issuance destinations and sources from document images face challenges due to variability in document layouts, leading to increased costs and time requirements for training models, as they struggle to accurately distinguish between billing and issuance destinations/sources.
Innovation Solution
The proposed solution involves generating a document type estimation model and an item value estimation model using machine learning, which determines the document type and candidate character strings, and then applies these to a rule-based algorithm to accurately extract the issuance destination and source character strings, reducing the burden of extensive training data and costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a trained model is generated to estimate character strings such as company names of issuance destination and source by learning the correspondence of character strings, then the extraction accuracy may be improved, but the training data requirement increases significantly and the cost and time for generating the trained model increases
Solution Approach 1:
The patent segments the extraction process into two independent stages: first extracting candidate character strings using a trained model, then determining issuance destination/source using rule-based algorithms. This segmentation allows the trained model to focus only on candidate extraction without needing to learn complex issuance role relationships, reducing training data requirements while maintaining accuracy.
Solution Approach 2:
The patent introduces rule-based algorithms as an intermediary between the trained model and final extraction results. The trained model generates candidate character strings, which are then processed by rule-based algorithms that apply document type and position information to determine issuance destination and source. This intermediary approach avoids the need for extensive training data covering all possible issuance scenarios.
2Measurement precision
If a trained model is generated to estimate character strings such as company names of issuance destination and source by learning the correspondence of character strings, then the extraction accuracy may be improved, but the cost for generating the trained model increases
Solution Approach 1:
The patent segments the extraction process into two independent stages: first extracting candidate character strings using a trained model, then determining issuance destination/source using rule-based algorithms. This segmentation allows the trained model to focus only on candidate extraction without needing to learn complex issuance role relationships, reducing training data requirements while maintaining accuracy.
Solution Approach 2:
The patent introduces rule-based algorithms as an intermediary between the trained model and final extraction results. The trained model generates candidate character strings, which are then processed by rule-based algorithms that apply document type and position information to determine issuance destination and source. This intermediary approach avoids the need for extensive training data covering all possible issuance scenarios.
3Measurement precision
If a trained model is generated to estimate character strings such as company names of issuance destination and source by learning the correspondence of character strings, then the extraction accuracy may be improved, but the training data requirement increases
Solution Approach 1:
The patent segments the extraction process into two independent stages: first extracting candidate character strings using a trained model, then determining issuance destination/source using rule-based algorithms. This segmentation allows the trained model to focus only on candidate extraction without needing to learn complex issuance role relationships, reducing training data requirements while maintaining accuracy.
Solution Approach 2:
The patent introduces rule-based algorithms as an intermediary between the trained model and final extraction results. The trained model generates candidate character strings, which are then processed by rule-based algorithms that apply document type and position information to determine issuance destination and source. This intermediary approach avoids the need for extensive training data covering all possible issuance scenarios.
Data Source
AI summary
Provided is an information processing apparatus including: an obtaining unit configured to obtain a token string generated based on character strings included in a document image; a first determination unit configured to determine a document type represented by the document image and character strings corresponding to a first item included in the document image by using a result obtained by inputting the token string into a trained model; and a second determination unit configured to determine a character string corresponding to a second item by applying the document type and the character strings corresponding to the first item to a rule-based algorithm.