Leveraging ShuffleNet transfer learning to enhance handwritten character recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Handwritten character recognition is crucial for applications like assistive technology and document processing.
• Accurate conversion of handwriting to digital formats is essential for computer-based systems.
• Existing methods face challenges in achieving high accuracy and efficiency for diverse handwriting styles.

Purpose of the Study:

• To propose an accurate and precise autonomous system for offline handwritten character and number recognition.
• To leverage transfer learning with ShuffleNet convolutional neural networks (CNNs) for multi-class recognition.
• To evaluate the system's performance against state-of-the-art character recognition methods.

Main Methods:

• Utilized a ShuffleNet CNN architecture for developing the handwriting recognition system.
• Employed transfer learning to train, validate, and recognize handwritten character and digit datasets.
• Categorized data into 26 classes for English characters and 10 classes for digits.

Main Results:

• Achieved an exceptional overall recognition accuracy of 99.50%.
• Demonstrated superior performance compared to other contrasted character recognition systems.
• Recorded a low computational cost with an average inference time of 2.7 ms per sample.

Conclusions:

• The proposed ShuffleNet-based system offers a highly accurate and efficient solution for offline handwritten character recognition.
• The system's high accuracy and low computational cost make it suitable for real-world applications.
• Transfer learning with ShuffleNet CNN is effective for multi-class handwritten character recognition tasks.