Multi-CNN Voting Method for Improved Arabic Handwritten Digits Classification
Areeg Fahad Rasheed, M. Zarkoosh, Sana Sabah Al-Azzawi · 2023
Handwritten Arabic recognition poses a unique set of challenges due to the intricate nature of the script, its diverse styles, and the inherent complexity of the Arabic language. With the increasing need for automated recognition systems in various domains, such as document analysis, postal services, and historical document preservation, finding the best system for accurate and efficient recognition becomes crucial. The primary objective of this paper is to propose a method for enhancing the performance of handwritten Arabic written classification tasks by incorporating ensemble learning that employs multiple CNN classical networks, including Inception, Lenet-5, AlexNet, Dense, MobileV2, and ResNet. The final r esults a re t hen c hosen using a voting approach. To validate the effectiveness of our proposed method, we evaluated it on the MADBase dataset and compared it with the latest works in the field. The results demonstrate that our proposed method achieved better accuracy, Fl-score, precision, and recall, with 99.31, 99.31, 99.31, and 99.30, respectively, outperforming existing research. This proposed method can be adapted and utilized for various classification tasks.1.