Abstract:
Human being has the ability to communicate and transfer information using a complex set of words, structures and grammar. They not only communicate and share the information but also, they have developed means to store and save their knowledge. Nowadays, digitization of the available information/knowledge is the best way to secure and transmit info. Softwares are programmed to analyze data and make automatic decisions. Among various machine learning techniques, CNN is designed to recognize characteristic features of handwritten digits. It is need of the time to convert all the available data (novels, journals, newspapers etc.) to computer recognizable texts. A standard and comprehensive dataset on majority of cursive languages is limited. Pashto is one of them. It is an important and National language of Afghanistan, as well as its the major language spoken in Pakistan (northern areas i-e. Khyber Pakhtunkhwa). This work proposes an optimized version of the VGG network, VGG-17, for Pashto handwritten character recognition. The trained VGG networks i.e VGG-16 and 19 and also propped an optimized VGG network i.e VGG-17. The VGG-17 achieved a high accuracy of 75% as compared to state-of-the-art image classification models VGG-16 and VGG-19 which achieved 70 and 72 accuracy respectively. The VGG-17 network also outperformed VGG-16 and VGG-19 by achieving high precision, recall, and f1 score values. The research shows that Pashto's handwritten characters can be better recognized by an optimized version of VGG-19 that is lightweight, easy to train, and highly accurate.
Page(s):
1-1
DOI:
DOI not available
Published:
Journal: Second International Conference on Computing Technologies, Tools and Applications (ICTAPP-24), June 4-6,2024 (Abstract Book), Volume: 0, Issue: 0, Year: 2024
Keywords:
Convolutional Neural Network CNN
,
Pashto Handwritten Character Recognition
,
Visual Geometry Group VGG