Classification of Handwriting Impairment Using CNN for Potential Dyslexia Symptom

Abstract

Early detection of symptoms is very important to help dyslexic children because they do not imply low intelligence. If dyslexic children are not assisted at an early stage, they will be left behind in education by their peers. Therefore, this project is helpful for diagnosing dyslexia symptoms by detecting handwriting impairment at early detection using machine learning. Dyslexia can occur in all languages but usually dyslexia in other than non-letter such as Chinese characters is lack focusing due to different handwriting characters. This study is focusing on processing Chinese character handwriting images to classify the potential dyslexia symptoms. The classification of potential dyslexia symptom is classified into 4 classes which Normal, Radical Error, Radical-Structure Error and Structure Error. The image augmentation is used to improve the performance of CNN based on in terms of its accuracy and precision. Thus, the accuracy of the training performance classification is 95.66%, while the accuracy of the validation is 96.20%.