政大學術集成


Please use this identifier to cite or link to this item: https://ah.nccu.edu.tw/handle/140.119/135535


Title: Toward Automatic Recognition of Cursive Chinese Calligraphy : An Open Dataset For Cursive Chinese Calligraphy Text
Authors: 廖文宏
Liao, Wen-Hung
Liang, Jung
Wu, Yi-Chieh
Contributors: 資科系
Keywords: Cursive Chinese Calligraphy , Text Recognition , Deep Learning
Date: 2020-01
Issue Date: 2021-06-04 14:49:11 (UTC+8)
Abstract: Calligraphy is one of the most important writing tools as well as cultural heritage in ancient China. Compared with other calligraphy styles, the cursive script is least restricted and oftentimes exhibits the personality of calligraphers. However, this style-oriented expression makes the cursive script hard to recognize even for trained experts. The call for auxiliary tools for cursive Chinese calligraphy text recognition has thus arisen.Data play a key role in the era of deep learning, yet there is a lack of open databases for the cursive Chinese calligraphy. In this paper, we address this discrepancy by collecting 43000 images consisting of 5301 different cursive Chinese calligraphy text. We have augmented the database with basic image processing operations to obtain a training set containing a total of 656K images. After experimenting with several deep neural architectures, we provided a baseline model Enhanced M6 (EM6) as a proof-of-concept to tackle the classification task. The proposed EM6 model achieved 60.3% top-1 accuracy and 80.8% top-5 accuracy on the evaluation data set, an indication that deep neural network has the potential to undertake the mission of cursive calligraphy recognition.
Relation: Proceedings of 2020 14th International Conference on Ubiquitous Information Management and Communication (IMCOM), IEEE SMC Society, pp.1-5
Data Type: conference
DOI link: https://doi.org/10.1109/IMCOM48794.2020.9001777
Appears in Collections:[Department of Computer Science ] Proceedings

Files in This Item:

File Description SizeFormat
292.pdf1464KbAdobe PDF32View/Open


All items in 學術集成 are protected by copyright, with all rights reserved.


社群 sharing