用深度卷积网络识别手写天城文字符，准确率达96.36%
Devanagari Handwritten Character Recognition using Convolutional Neural Network
- 构建两层卷积神经网络，针对天城文手写字符设计识别模型
- 在36类字符数据集上测试准确率达96.36%，训练准确率99.55%
- 适合从事印度语手写识别或中文之外亚洲文字研究者参考
手写字符识别因在搜索引擎、社交媒体、推荐系统等场景的应用而受到关注。天城文是印度最古老的书写系统之一，缺乏完善的数字化工具。本文提出一种自动化方法，从天城文图像中提取手写梵文字符，以节省时间并处理陈旧数据。采用两层深度卷积神经网络，基于开放的天城文手写字符数据集（DHCD）进行训练与测试，该数据集包含36个类别，每类1700张图像。实验结果显示，模型在测试集上达到96.36%的识别准确率，在训练集上高达99.55%。
原文摘要 · Abstract (English)
Handwritten character recognition is getting popular among researchers because of its possible applications in facilitating technological search engines, social media, recommender systems, etc. The Devanagari script is one of the oldest language scripts in India that does not have proper digitization tools. With the advancement of computing and technology, the task of this research is to extract handwritten Hindi characters from an image of Devanagari script with an automated approach to save time and obsolete data. In this paper, we present a technique to recognize handwritten Devanagari characters using two deep convolutional neural network layers. This work employs a methodology that is useful to enhance the recognition rate and configures a convolutional neural network for effective Devanagari handwritten text recognition (DHTR). This approach uses the Devanagari handwritten character dataset (DHCD), an open dataset with 36 classes of Devanagari characters. Each of these classes has 1700 images for training and testing purposes. This approach obtains promising results in terms of accuracy by achieving 96.36% accuracy in testing and 99.55% in training time.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。