Related Experiment Video
Updated: Oct 13, 2025
08:08
Comparing the Frequency Effect Between the Lexical Decision and Naming Tasks in Chinese
Published on: April 1, 2016
9.5K
FAN-MCCD: Fast and Accurate Network for Multi-Scale Chinese Character Detection
1Department of Electronics Engineering, Yeungnam University, 280 Daehak-ro, Gyeongsan-si 38541, Korea.
Sensors (Basel, Switzerland)
|November 13, 2021
Summary
This study introduces an efficient method for accurately detecting handwritten Chinese characters in historical documents, significantly improving detection rates and reducing errors compared to existing models.
Area of Science:
- Computer Vision
- Document Analysis
- Artificial Intelligence
Background:
- Scale variation in character detection leads to inaccurate localization, particularly in historical and handwritten documents.
- Overconfidence in document analysis results is a significant issue within the research community.
Purpose of the Study:
- To explore the performance of a state-of-the-art network for fast and accurate handwritten Chinese character prediction in old documents.
- To adapt character detection to multi-scale locations precisely, minimizing pre-processing steps.
Main Methods:
- Utilized a neural network with multi-scale feature maps to adapt to character locations.
- Employed pre-selected bounding boxes of varying scales and aspect ratios across feature maps.
- Applied non-maximum suppression to prune bounding boxes for final results.
Main Results:
- Achieved enhanced detection performance with a detection rate (DT) of 98.84%, false positive per character (FPPC) of 0.71, and F-score of 97.64% on Caoshu, Character, and Src-images datasets.
- Demonstrated superior performance compared to the Single-Shot Detector (SSD) model.
Conclusions:
- The proposed method significantly enhances handwritten Chinese character detection in historical documents.
- The optimized neural network architecture and loss function contribute to improved detection accuracy and efficiency.
