Related Experiment Video
Updated: Oct 29, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
728
Full depth CNN classifier for handwritten and license plate characters recognition
Mohammed Salemdeeb1, Sarp Ertürk2
1Department of Electrical-Electronics Engineering, Bartin University, Bartin, Turkey.
Peerj. Computer Science
|July 9, 2021
Summary
A novel deep learning model using stacked convolutional neural networks (CNNs) achieves high accuracy in recognizing Latin and Arabic handwritten characters and license plate data. This low-complexity, fast, and reliable architecture offers promising advancements for character recognition applications.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Character recognition is vital for numerous applications, with deep learning, particularly convolutional neural networks (CNNs), showing significant advancements.
- CNNs remain the state-of-the-art for image classification tasks, including character recognition.
Purpose of the Study:
- To propose a novel, simple, full-depth stacked CNN architecture for recognizing Latin and Arabic handwritten alphanumeric characters.
- To adapt and evaluate this architecture for license plate (LP) character recognition.
Main Methods:
- The proposed architecture comprises four convolutional layers, two max-pooling layers, and one fully connected layer.
- The model was tested on diverse datasets including handwritten characters (MNIST, Fashion-MNIST, MAHDB, AHCD, AIA9K) and license plate characters (real-world isolated characters from Saudi Arabia, Turkey, Europe, USA, UAE, KSA).
Main Results:
- The CNN architecture demonstrated high accuracy across multiple datasets, achieving error rates as low as 0.28% for MNIST and 0.26% for Saudi license plates.
- Excellent performance was observed on various handwritten and license plate character datasets, indicating the model's robustness and effectiveness.
Conclusions:
- The developed CNN architecture is low-complex, fast, reliable, and achieves high classification accuracy.
- This approach shows potential to advance the field of character recognition by offering a balance of low complexity, high accuracy, and comprehensive feature extraction.
