OCR False Recognition Correction via Character Association
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current document processing systems for correcting character recognition errors in OCR processes lack efficiency in extracting specific characters or character strings, particularly when characters are likely to be recognized falsely, as they do not effectively handle variations and false recognition patterns.
Innovation Solution
An information processing apparatus with a recognition unit, memory, and extraction unit that associates characters with varying recognition accuracy, allowing for the registration and extraction of specific character strings by replacing falsely recognized characters with their correct counterparts, using a false-recognition dictionary and association dictionary to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If OCR recognition is performed on character strings containing characters likely to be falsely recognized, then the extraction process can handle more varied input, but the recognition accuracy decreases
Solution Approach 1:
The system performs preliminary actions by pre-registering extraction targets and their corresponding false recognition patterns in a dictionary before the actual extraction process. When a character is recognized, the system checks the recognition result against the stored false recognition patterns and automatically corrects matches, thereby maintaining high accuracy while handling varied input characters.
Solution Approach 2:
The system implements feedback by comparing OCR recognition results with stored false recognition patterns and automatically correcting identified errors. The recognition unit outputs character information, the extraction unit compares this against the dictionary of known false patterns, and corrections are applied based on this feedback loop, improving accuracy for characters prone to false recognition.
2Measurement precision
If a dictionary-based correction system is implemented, then recognition accuracy improves, but device complexity increases
Solution Approach 1:
The memory unit serves multiple functions: it stores both the dictionary of false recognition patterns and the extraction target information, acting as both a reference database and a storage mechanism. This multi-functionality reduces the need for separate correction modules, thereby limiting the increase in system complexity while maintaining improved recognition accuracy.
Solution Approach 2:
The extraction unit performs self-correction by autonomously comparing recognition results against the stored false recognition patterns and applying corrections without requiring external intervention. The system serves its own correction needs using internally stored data, reducing the complexity that would arise from adding separate correction apparatus.
3Productivity
If multiple extraction targets including false recognition patterns are registered, then extraction completeness improves, but processing time increases
Solution Approach 1:
The system performs preliminary action by pre-registering all extraction targets including false recognition patterns in the memory during system initialization or setup phases. This allows the extraction unit to perform rapid lookups during actual extraction operations, improving completeness while minimizing processing time during runtime since the comparison work has been prepared in advance.
Data Source
AI summary
An information processing apparatus includes a recognition unit, a memory, a registration unit, and an extraction unit. The recognition unit performs recognition of one or more characters included in an image and outputs character information indicating the one or more characters. The memory stores a first specific character and a second specific character in association with each other. The first specific character has an index taking on a value lower than a predetermined value. The index indicates accuracy of the recognition performed by the recognition unit. The second specific character is output after the recognition unit performs the recognition of the first specific character. The registration unit registers at least one character or at least one character string as an extraction target. The registration unit registers a first character or a first character string as an extraction target and a second character or a second character string as an extraction target if an operator performs an operation for registering the first character or the first character string as the extraction target and if the first character or the first character string includes the first specific character. The second character or the second character string is composed by replacing the first specific character in the first character or the first character string with the second specific character. The extraction unit extracts at least one character or at least one character string from the character information. The extraction unit extracts the first character or the first character string and the second character or the second character string that are registered as the respective extraction targets.