Label Range Integration for CTC Recognition Outputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing series label recognition technologies, such as CTC, fail to explicitly separate label boundaries, making it difficult to determine the ranges of labels corresponding to recognition results.
Innovation Solution
An information processing device comprising a feature quantity acquiring unit, range estimating unit, associating unit, and integrating unit, which estimates and integrates ranges of data associated with labels using neural networks to specify the ranges of labels in a label string.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If series label recognition technology such as CTC is used to achieve high recognition accuracy, then recognition accuracy is improved, but the ability to specify label ranges deteriorates
Solution Approach 1:
The patent divides the label string into individual label units, each with associated range information. By segmenting the continuous label sequence and attaching range data to each segment, the system preserves both the high accuracy of CTC recognition and the ability to specify individual label ranges.
Solution Approach 2:
The patent introduces an intermediary data structure that connects the CTC recognition output with range information. This intermediary layer allows the system to maintain the simplicity and accuracy of CTC while adding the capability to track and specify label ranges without disrupting the original recognition process.
2Ease of operation
If conventional methods are used to estimate character regions from images, then region estimation is possible, but consistency with CTC recognition technology deteriorates
Solution Approach 1:
The patent merges the region estimation function with the CTC recognition framework by integrating range information extraction into the same neural network architecture. This unification ensures that both CTC recognition and range estimation operate consistently within a single methodological framework, eliminating the inconsistency between separate approaches.
Solution Approach 2:
The patent creates a universal system that performs both CTC recognition and range estimation through a single integrated approach. The same neural network and processing framework that enables high-accuracy recognition also provides consistent range information, making the system multi-functional while maintaining methodological consistency.
Data Source
AI summary
An information processing device includes a feature quantity acquiring unit, a range estimating unit, an associating unit, an integrating unit, and an output unit. The feature quantity acquiring unit acquires a feature quantity extracted from data consisting of a plurality of values. The range estimating unit estimates a range of the data in which an element to which a predetermined label is to be assigned may be present from the acquired feature quantity. The associating unit associates each label with at least one of a plurality of feature quantities. The integrating unit performs integration processing of integrating one or more ranges of the data estimated from each of one or more feature quantities that are associated with the label into one range of the data. The output unit outputs a correspondence relationship between the label and the range of the data that has been subjected to the integration processing.