Dysgraphia Handwriting Image Augmentation for CNN Model Classification

• Published in: 14th IEEE International Conference on Control System, Computing and Engineering, ICCSCE 2024 - Proceedings
• Main Author: Harun N.; Isa I.S.; Ramlan S.A.; Osman M.K.; Maruzuki M.I.F.
• Format: Conference paper
• Language: English
• Published: Institute of Electrical and Electronics Engineers Inc. 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85207057259&doi=10.1109%2fICCSCE61582.2024.10696383&partnerID=40&md5=38e6b594ddb991945858b6b84d299b4d

Dysgraphia affects a person's ability to write consistently and properly especially among school children. It is a challenging condition as it needs effective intervention to help the affected children succeed academically and socially. With the advancement in technology in artificial intelligence (AI), various methods and approaches have been developed using convolutional neural networks (CNN) model to overcome several limitations to assess dysgraphia symptoms. However, there are major concerns about the difficulties of getting large data of dysgraphia handwriting images for CNN attributes model. Thus, this study is aimed to develop dysgraphia handwriting recognition model based on augmentation method. In this study, image augmentation is addressed by creating new data by using rotation and brightness technique to generate a set of synthetic images. The augmented data is trained and tested using CNN classification model to classify four classes of dysgraphia handwriting. The results show a significant improvement with 77% accuracy using augmented as compared to without augmented data only 73%. This study indicated that augmentation method is significant for inclusion in CNN classification model particularly for dysgraphia potential risk recognition. This study is further recommended to implement intelligence-based augmentation method which can be incorporated into a computer-assisted dysgraphia screening system to provide a rapid, accurate, and unbiased dysgraphia detection. © 2024 IEEE.