Deep Learning-Based Child Handwritten Arabic Character Recognition and Handwriting Discrimination

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Existing handwriting recognition systems primarily focus on adult data.
• Child handwriting presents unique challenges due to lower quality, higher variation, and distortions.
• Current systems are often not optimized for recognizing children's handwriting.

Purpose of the Study:

• To develop a novel convolution neural network (CNN) model for recognizing children's handwritten isolated Arabic letters.
• To investigate the impact of training data composition (child, adult, or both) on recognition accuracy.
• To enhance writer-group classification by combining deep features with supplementary features.

Main Methods:

• Development of a new CNN model for Arabic character recognition.
• Experimental evaluation using datasets of children's (Hijja) and adults' (AHCD) handwriting.
• Integration of supplementary features with CNN-extracted features for improved classification.
• Comparison of performance using classifiers like Softmax, SVM, KNN, and Random Forest.

Main Results:

• Training strategies significantly influence recognition performance.
• Including adult data in the training set improved child handwritten character recognition accuracy to approximately 93%.
• Fusion of supplementary and deep features enhanced child handwriting discrimination accuracy to around 94%.

Conclusions:

• The proposed CNN model demonstrates effectiveness in recognizing children's handwritten Arabic characters.
• Adult data inclusion is crucial for improving the performance of child handwriting recognition systems.
• Combining deep and supplementary features offers a promising approach for distinguishing between child and adult handwriting.