An intelligent approach for Arabic handwritten letter recognition using convolutional neural network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Digital transformation is crucial for businesses, but digitizing manual documentation, especially handwriting, presents significant challenges.
• Arabic handwriting recognition is particularly complex due to script characteristics and handwriting variations.

Purpose of the Study:

• To develop an intelligent approach for recognizing handwritten Arabic letters.
• To address the complexities of Arabic handwriting digitization for improved business process automation.

Main Methods:

• A convolution neural network (CNN) model was designed for handwritten Arabic letter recognition.
• The CNN model incorporated batch normalization and dropout regularization to prevent overfitting.
• The model was evaluated on the Arabic Handwritten Characters Dataset (AHCD) comprising 16,800 letters.

Main Results:

• The proposed CNN model achieved a high accuracy of 96.78% in recognizing handwritten Arabic letters.
• Dropout regularization significantly improved model performance and prevented overfitting.
• The model demonstrated superior performance compared to existing domain-relevant approaches.

Conclusions:

• The intelligent CNN-based approach offers an effective solution for handwritten Arabic letter recognition.
• This method facilitates the digital transformation of manual Arabic documents, enhancing efficiency.
• The study highlights the importance of regularization techniques in deep learning for complex pattern recognition tasks.