Document Image Extraction Using Machine-Learned Attribute Groups
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to accurately extract item values from document images when they are noted by irregular character strings, such as personal names or addresses, due to the difficulty in evaluating validity using data types.
Innovation Solution
An information processing apparatus that utilizes machine learning to identify attribute groups for character strings, converting them into feature vectors, and extracts candidate character strings matching a predetermined extraction target by employing a character string classifier.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional methods using predetermined keywords and data types are used to extract item values, then extraction is successful for regular character strings, but extraction fails for irregular character strings such as personal names, addresses, occupations, or relationships
Solution Approach 1:
The patent changes the parameter of character string evaluation from fixed data types to dynamic attribute groups. By using machine learning to identify attributes such as personal name, address, occupation, and relationship, the system adapts to irregular character strings while maintaining extraction accuracy. This resolves the contradiction by making the extraction method versatile without sacrificing reliability.
Solution Approach 2:
The patent introduces an intermediary mechanism (attribute group identification based on machine learning) between the character string and the extraction process. This intermediary enables the system to handle both regular and irregular character strings by classifying them into appropriate attribute groups, thereby improving both reliability and adaptability simultaneously.
2Reliability
If attribute group identification using machine learning is implemented, then extraction accuracy for irregular character strings improves, but system complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the machine learning model to identify attribute groups before the actual extraction process. This preliminary training phase enables the system to automatically classify character strings into appropriate attribute groups, reducing the complexity of the real-time extraction process while maintaining high accuracy for irregular character strings.
3Productivity
If data type-based validity evaluation is used, then processing is simple and fast, but irregular character strings cannot be properly evaluated
Solution Approach 1:
The patent applies partial action by using machine learning-based attribute group identification only for character strings that require it (irregular ones), while maintaining simple data type-based evaluation for regular character strings. This selective approach preserves processing speed for common cases while improving validity evaluation accuracy for irregular cases, resolving the contradiction between productivity and reliability.
Data Source
AI summary
Provided is an information processing apparatus that extracts one or more candidate character strings to be a candidate of an item value corresponding to a predetermined item, from among multiple character strings included in a document image to be processed. The information processing apparatus first obtains information indicating an attribute group of a preset extraction target, then identifies an attribute group to which each of the multiple character strings belongs from among predetermined multiple attribute groups based on a feature vector corresponding to each of the multiple character strings included in the document image to be processed, and subsequently extracts the character string whose identified attribute group is the same as the attribute group of the extraction target from among the multiple character strings included in the document image to be processed, as the candidate character string.