Ensemble deep transfer learning model for Arabic (Indian) handwritten digit recognition

Recognising handwritten digits or characters is a challenging task due to noisy data that results from different writing styles. Numerous applications essentially motivate to build an effective recognising model for such purposes by utilizing recent intelligent techniques. However, the difficulty em...

Full description

Saved in:
Bibliographic Details
Published in:Neural computing & applications 2022, Vol.34 (1), p.705-719
Main Authors: Alkhawaldeh, Rami S., Alawida, Moatsum, Alshdaifat, Nawaf Farhan Funkur, Alma’aitah, Wafa’ Za’al, Almasri, Ammar
Format: Article
Language:English
Subjects:
Arabic language
Architecture
Artificial Intelligence
Artificial neural networks
Character recognition
Computational Biology/Bioinformatics
Computational Science and Engineering
Computer Science
Data Mining and Knowledge Discovery
Datasets
Deep learning
Digits
Feature extraction
Handwriting recognition
Image Processing and Computer Vision
Information systems
Machine learning
Model accuracy
Neural networks
Optimization techniques
Original Article
Probability and Statistics in Computer Science
Writers
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recognising handwritten digits or characters is a challenging task due to noisy data that results from different writing styles. Numerous applications essentially motivate to build an effective recognising model for such purposes by utilizing recent intelligent techniques. However, the difficulty emerges when using the Arabic language that suffers from diverse noises; because of the way of writing inherent in connecting characters and digits. Therefore, this work focuses on the Arabic (Indian) digits and propose an ensemble deep transfer learning (EDTL) model that efficaciously detect and recognise these digits. The EDTL model is a combination of two effective pre-trained transfer learning models that consume time and cost complexity in the training phase. The EDTL is trained on large datasets to extract relevant features as input to a fully-connected Artificial Neural Network classifier. The experimental results, using popular datasets, show significant performance obtained by the EDTL model with accuracy reached up to 99.83% in comparison to baseline methods include deep transfer learning models, ensemble deep transfer learning models and state-of-the-art techniques.
ISSN:0941-0643
1433-3058
DOI:10.1007/s00521-021-06423-7