Open Access iconOpen Access

ARTICLE

Letter Recognition Reinvented: A Dual Approach with MLP Neural Network and Anomaly Detection

Nesreen M. Alharbi*, Ahmed Hamza Osman, Arwa A. Mashat, Hasan J. Alyamani

Faculty of Computing and Information Technology in Rabigh, King Abdulaziz University, Jeddah, 21589, Saudi Arabia

* Corresponding Author: Nesreen M. Alharbi. Email: email

Computer Systems Science and Engineering 2024, 48(1), 175-198. https://doi.org/10.32604/csse.2023.041044

Abstract

Recent years have witnessed significant advancements in the field of character recognition, thanks to the revolutionary introduction of machine learning techniques. Among various types of character recognition, offline Handwritten Character Recognition (HCR) is comparatively more challenging as it lacks temporal information, such as stroke count and direction, ink pressure, and unexpected handwriting variability. These issues contribute to a poor level of precision, which calls for the adoption of anomaly detection techniques to enhance Optical Character Recognition (OCR) schemes. Previous studies have not researched unsupervised anomaly detection using MLP for handwriting recognition. Therefore, this study proposes a novel approach for enhanced English letter identification based on a Multi-Layer Perceptron (MLP) neural network and an anomaly detection method. The MLP neural network is trained on a massive dataset of letters, enabling it to achieve high accuracy in recognizing letters from different font styles. Additionally, to further enhance the performance of the system, an anomaly detection method is employed to identify and eliminate any outliers or anomalies in the recognition process. The combination of these two techniques results in a highly accurate and robust letter recognition system that can be applied to a wide range of practical applications. The hybrid approach aims to increase the accuracy of English letter identification while minimizing misclassification, i.e., classification issues associated with recognizing English letters in images. To identify handwritten letters based on their underlying patterns, discriminant samples with extracted attributes were fed into a neural network, which was then used to categorize the data. This paper achieved significant improvements in English letter identification by coupling the anomaly detection algorithm and the multilayer perceptron neural network. When tested on the UCI-English letters dataset, this hybrid technique boosted the accuracy from 84.11% using the multiple neural networks to an impressive 92.84% with the proposed method. Moreover, the experimental results of the study have demonstrated that the hybrid method outperforms modern classification techniques for handwriting letter recognition. Overall, the proposed approach has tremendous potential to advance the field of character recognition, especially for offline handwritten character recognition, and can pave the way for developing more accurate and efficient OCR systems in the future.

Keywords


Cite This Article

APA Style
Alharbi, N.M., Osman, A.H., Mashat, A.A., Alyamani, H.J. (2024). Letter recognition reinvented: A dual approach with MLP neural network and anomaly detection. Computer Systems Science and Engineering, 48(1), 175-198. https://doi.org/10.32604/csse.2023.041044
Vancouver Style
Alharbi NM, Osman AH, Mashat AA, Alyamani HJ. Letter recognition reinvented: A dual approach with MLP neural network and anomaly detection. Comput Syst Sci Eng. 2024;48(1):175-198 https://doi.org/10.32604/csse.2023.041044
IEEE Style
N.M. Alharbi, A.H. Osman, A.A. Mashat, and H.J. Alyamani, “Letter Recognition Reinvented: A Dual Approach with MLP Neural Network and Anomaly Detection,” Comput. Syst. Sci. Eng., vol. 48, no. 1, pp. 175-198, 2024. https://doi.org/10.32604/csse.2023.041044



cc Copyright © 2024 The Author(s). Published by Tech Science Press.
This work is licensed under a Creative Commons Attribution 4.0 International License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
  • 527

    View

  • 304

    Download

  • 0

    Like

Share Link