Convolutional ensembles for Arabic Handwritten Character and Digit Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Arabic handwritten character and digit recognition is a challenging task due to script variations.
• Existing methods often struggle with achieving high accuracy on diverse datasets.

Purpose of the Study:

• To develop and validate a robust learning algorithm for Arabic Handwritten Character and Digit (AHCD) recognition.
• To achieve state-of-the-art classification accuracies on benchmark datasets.

Main Methods:

• An ensemble architecture of Convolutional Neural Networks (CNNs) was employed.
• A hybrid training algorithm combining adaptive and stochastic gradient descent was utilized.
• Monte Carlo Cross-Validation and K-fold Cross-Validation were used for robust evaluation.
• Hyper-parameter tuning was performed using the MADbase digits dataset.

Main Results:

• Achieved average validation and testing classification accuracies of 99.74% and 99.47% on the MADbase digits dataset.
• Attained state-of-the-art validation and testing accuracies of 98.60% and 98.42% on the AHCD character dataset.
• Demonstrated superior performance in recognizing Arabic handwritten characters and digits.

Conclusions:

• The proposed learning algorithm and CNN ensemble architecture are highly effective for Arabic handwritten character and digit recognition.
• The hybrid training strategy facilitates convergence and improves classification accuracy.
• The results represent a significant advancement in the field of optical character recognition for Arabic script.