Image Processing Apparatus for Accurate Legend-Based Value Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods fail to accurately extract user-selected values from scanned documents, particularly when legends like '1: MALE 2: FEMALE' are used, as they cannot identify the area for the legend or obtain the corresponding value from the scanned image.
Innovation Solution
An image processing system that extracts item names and input information, and when the input is unsuitable, uses legend information to determine the corresponding value by searching for and matching legend formats within the scanned image, allowing for accurate value extraction even when the legend and value areas differ.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the area in which the legend is entered is used to identify the value, then the value can be obtained, but the method cannot identify the area when the legend and value are entered in different areas
Solution Approach 1:
The patent segments the document image into multiple regions including a legend region and a value entry region. By dividing the document structure into distinct segments, the system can independently analyze and identify the legend area separately from the value entry area, enabling accurate value extraction even when they are positioned in different locations.
Solution Approach 2:
The patent introduces an intermediary process that establishes the correspondence relationship between the legend and the value entry area. This intermediary mechanism links the legend region to the appropriate value region through spatial relationships or document structure analysis, enabling the system to correctly identify which value corresponds to which legend option.
2Productivity
If pre-stored business-form formats and conversion rules are used, then value extraction can be performed, but development costs and system complexity increase
Solution Approach 1:
The patent enables the system to automatically identify and analyze the document structure, legend positions, and value entry areas without requiring pre-stored formats or manual configuration. The system serves itself by dynamically adapting to different document layouts through image processing and spatial relationship analysis, eliminating the need for extensive pre-programming of specific form formats.
Solution Approach 2:
The patent changes the approach from using fixed pre-stored formats to dynamically analyzing document parameters such as spatial relationships, text patterns, and structural features. By adjusting to varying document parameters rather than relying on fixed templates, the system maintains simplicity while achieving broad applicability across different document types.
Data Source
AI summary
A technique in the present disclosure makes it possible to accurately obtain a value selected by a user with respect to a selection-type item for which a legend is prepared in a situation where a value entered in a document is extracted from a scanned image of the document. An item name is extracted from a document image, input information input by a user is extracted from the document image, and a legend is extracted from the document image. Additionally, in a case where the input information is unsuitable as a value corresponding to the item name, the value corresponding to the item name is obtained based on the legend and the input information.