OCR Key Value Extraction Using Attribute Mediation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OCR technologies face difficulties in extracting key character strings due to erratic recognition or their absence in documents, making it challenging to output key and value character strings effectively.
Innovation Solution
An information processing apparatus equipped with a processor that acquires attribute information from character recognition results to identify and output key and value character strings, even when the key character string is difficult to extract or not present, by utilizing a learning model to determine attributes and types of character strings and extracting corresponding key strings based on positional relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If OCR operation is performed to extract character strings from images, then character recognition is achieved, but key character strings may be difficult to extract due to erratic recognition or absence in documents
Solution Approach 1:
The patent introduces an intermediary approach by using attribute information as a mediator between the OCR recognition results and the key character string extraction. Instead of directly extracting key character strings from OCR results, the system first extracts attribute information (such as text type, position, layout properties) from the recognition results, then uses this attribute information to reliably identify and extract key character strings and their corresponding value character strings, even when direct OCR extraction fails or is unreliable
2Productivity
If traditional key value extraction is performed, then value character strings can be extracted, but key character strings corresponding to value character strings may not be included in the document
Solution Approach 1:
The patent applies preliminary action by first extracting attribute information from all character strings in the document before attempting to identify key-value pairs. This preliminary extraction of attributes (such as text properties, positional relationships, layout characteristics) enables the system to subsequently identify which character strings serve as keys and which as values, even when the key character string is not explicitly present in the document. This preliminary analysis prevents information loss by establishing a framework for identification before the actual key-value pairing occurs
3Measurement precision
If attribute information is acquired and used to extract key character strings, then accurate extraction is achieved, but processing complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the character string extraction process into distinct stages: first extracting attribute information from character recognition results, then using this attribute information to identify key character strings, and finally extracting value character strings based on the identified keys. This segmentation transforms a complex single-step extraction problem into manageable sequential steps, where each step handles a specific aspect of the extraction task, reducing overall system complexity while maintaining high extraction accuracy
Data Source
AI summary
An information processing apparatus includes a processor configured to: acquire, from results of character recognition performed on a target image including character strings, attribute information indicating an attribute to which a key character string and a value character string belong, the key character string as a character string specified beforehand as a key and the value character string as a character string indicating a value corresponding to the key character string; acquire by using the attribute information the key character string corresponding to the value character string extracted from the results of the character recognition; and output the key character string and the value character string corresponding to the key character string.