TY - JOUR
T1 - Deep learning autoencoder approach
T2 - Automatic recognition of artistic Arabic calligraphy types
AU - Al-Hmouz, Rami
N1 - Funding Information:
This project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah, Saudi Arabia under grant no. (G/362/135/37). The authors, therefore, acknowledge with thanks DSR technical and financial support.
Publisher Copyright:
© 2020 University of Kuwait. All rights reserved.
PY - 2020
Y1 - 2020
N2 - Recognition of Arabic calligraphy types is a challenging problem. Difficulties include similarities among different types, overlap between letters, and letters that assume different shapes. In this study, a deep learning approach to recognizing artistic Arabic calligraphy types is presented. Autoencoder is a deep learning approach with the capability of reducing data dimensions in addition to extract features. Autoencoders can be stacked with several layers. The system is composed of three layers consisting of two encoder layers to extract features and one soft max layer for the recognition stage. The font can be recognized in a collective manner based on the words or segments that exist in the font images. The input of the system consists of individual words or segment images that compose the font image, and the output is the recognized font type. The approach was evaluated on local and public datasets, and the achieved recognition rates were 92.1% and 99.5%, respectively.
AB - Recognition of Arabic calligraphy types is a challenging problem. Difficulties include similarities among different types, overlap between letters, and letters that assume different shapes. In this study, a deep learning approach to recognizing artistic Arabic calligraphy types is presented. Autoencoder is a deep learning approach with the capability of reducing data dimensions in addition to extract features. Autoencoders can be stacked with several layers. The system is composed of three layers consisting of two encoder layers to extract features and one soft max layer for the recognition stage. The font can be recognized in a collective manner based on the words or segments that exist in the font images. The input of the system consists of individual words or segment images that compose the font image, and the output is the recognized font type. The approach was evaluated on local and public datasets, and the achieved recognition rates were 92.1% and 99.5%, respectively.
KW - Artistic Arabic calligraphy
KW - Autoencoder
KW - Deep learning
KW - Optical font recognition
UR - http://www.scopus.com/inward/record.url?scp=85092067772&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85092067772&partnerID=8YFLogxK
M3 - Article
AN - SCOPUS:85092067772
SN - 2307-4108
VL - 47
SP - 2
EP - 14
JO - Kuwait Journal of Science
JF - Kuwait Journal of Science
IS - 3
ER -