Related Experiment Videos
A multimodal learning framework for Arabic handwritten word recognition and future research directions
1Department of Computer Engineering, College of Computer Engineering and Science, Prince Mohammad Bin Fahd University, Khobar, Saudi Arabia. jalkhateeb@pmu.edu.sa.
Scientific Reports
|July 1, 2026
Summary
Arabic handwritten word recognition is challenging. A new framework combining deep learning (CNN) and probabilistic models (HMM) achieved 95.4% accuracy, significantly outperforming traditional methods.
Area of Science:
- Computer Science
- Artificial Intelligence
- Pattern Recognition
Background:
- Arabic handwritten word recognition is complex due to script characteristics and variability.
- Existing approaches like probabilistic models and deep learning have limitations.
- Integrating complementary strengths of different models is underexplored.
Purpose of the Study:
- To present a unified framework for Arabic handwritten word recognition.
- To integrate Hidden Markov Model (HMM) and Convolutional Neural Network (CNN) approaches.
- To evaluate the performance of this integrated framework.
Main Methods:
- Developed a unified recognition framework combining HMM and CNN.
- Utilized a dataset of 5000 Arabic handwritten word images (Libyan city names).
- Evaluated the framework under identical experimental conditions.
Main Results:
- The CNN-based model achieved 95.4% recognition accuracy.
- The HMM-based system achieved 83.62% recognition accuracy.
- Learned visual representations demonstrated significant efficacy.
Conclusions:
- Deep learning (CNN) significantly outperforms traditional HMM for this task.
- The integrated framework highlights the efficacy of learned visual representations.
- Future work should focus on sequence-aware deep architectures and language model integration for scalability.
