Text Line Classifier Generation Using Font Reservoirs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Chinese text region recognition in natural scenes faces challenges due to low accuracy and robustness, high computational resource consumption, and the need for extensive marked-up samples, especially with the complexity of Chinese characters and various fonts and styles.
Innovation Solution
A method and apparatus for generating text line classifiers using a terminal system font reservoir to create samples, extract features, and train models with a BP neural network, allowing for high applicability and efficiency in recognizing text regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If CNN trained classifiers are used to classify Chinese text regions, then classification accuracy is improved, but computational resource consumption increases excessively
Solution Approach 1:
The patent extracts and utilizes pre-stored marked-up samples containing text line features to train classifiers, rather than using full CNN models. This extraction approach maintains classification accuracy while significantly reducing computational resource consumption by using a more efficient feature-based classification method.
2Measurement precision
If a large number of marked-up samples are used for training, then classification accuracy is improved, but the effort and cost of sample preparation increases
Solution Approach 1:
The patent performs preliminary actions by pre-storing marked-up samples with text line features before actual classification tasks. These pre-prepared samples are reused across different applications, eliminating the need for repeated sample marking and reducing preparation time while maintaining classification accuracy.
Solution Approach 2:
The pre-stored marked-up samples serve multiple functions across different application scenarios. The same sample set can be used for training various classifiers targeting different languages or text types, reducing the need to create new samples for each application and significantly cutting preparation time and cost.
3Device complexity
If experience threshold based classification is used, then algorithm simplicity is improved, but recognition accuracy and robustness decrease
Solution Approach 1:
The patent changes the parameters used for classification from simple experience-based thresholds to features extracted from pre-stored marked-up samples. This parameter transformation maintains algorithm simplicity while significantly improving recognition accuracy and robustness by using data-driven features instead of fixed thresholds.
4Adaptability or versatility
If classifiers are trained for different application requirements, then adaptability is improved, but the need for new marked-up samples increases
Solution Approach 1:
The patent creates universal pre-stored marked-up samples that can serve multiple application requirements. The same sample set is used to train classifiers for different languages, text types, or application scenarios, eliminating the need to create new samples for each application while maintaining high adaptability.
Data Source
AI summary
A method of generating a text line classifier including generating text line samples by use of a present terminal system font reservoir. The method also includes extracting features from the text line samples and pre-stored marked-up samples. The method further includes training models by use of the extracted features to generate a text line classifier for recognizing text regions. With the system font reservoir being utilized for generating text line samples, the generated text line classifiers can target different scenes or different requirements for text region recognition with a high degree of applicability and wide application in addition to ease of implementation. Together with the combinational use of the marked up samples for extracting features from the text line samples, the generated text line classifiers provide for enhanced classification efficiency and accuracy.