A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm
The processing of automatically generated images of documents, which is a complex and expensive but necessary component of handwriting recognition, and which has always drawn the attention of engineers and scientists. The definition includes processing of an image in which handwritten letters or num...
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/44601/ http://umpir.ump.edu.my/id/eprint/44601/1/A%20sequential%20handwriting%20recognition%20model%20based%20on%20a%20dynamically%20configurable%20convolution%20recurrent%20neural%20network%20and%20hybrid%20salp%20swarm%20algorithm.pdf |
| _version_ | 1848827138725117952 |
|---|---|
| author | Ahmed Ali Mohammed, Al-saffar |
| author_facet | Ahmed Ali Mohammed, Al-saffar |
| author_sort | Ahmed Ali Mohammed, Al-saffar |
| building | UMP Institutional Repository |
| collection | Online Access |
| description | The processing of automatically generated images of documents, which is a complex and expensive but necessary component of handwriting recognition, and which has always drawn the attention of engineers and scientists. The definition includes processing of an image in which handwritten letters or numbers exist with the term handwriting recognition. The practical applications of handwriting recognition in many of the current real-world applications rely on the processing of sequential texts, where the language must be detected with efficiency, suggesting the development of a flexible handwriting recognition model. The purpose of this research is to finally describe an automated and flexible way of finding the most appropriate CRNN (Convolutional Recurrent Neural Network) architecture for predictive-sequence related tasks. This research present a dynamic configurator of the CRNN (DC-CRNN), geared for sequence learning in the context of handwriting recognition, inspired by bio-inspired approaches. The built DCCRNN is based on the Salp Swarm optimization Algorithm (SSA), a processor that given a particular dataset will find the best CRNN’s structure and hyperparameters. The present research offers a unique hybridization of SSA with the Late Acceptance Hill-Climbing (LAHC) to further strengthen the optimization process. Experiments were performed on two well-known datasets of Arabic and English handwriting, IAM and IFN/ENIT. The empirical results display that the proposed DC-CRNN and its SSA hybridization are capable of autonomously finding and selecting the best CRNN for a specific dataset. The experimental results indicate that the implementation outperformed other classic handcrafted CRNNs in terms of performance, accuracy, and overall quality for predictive sequential handwriting recognition tasks. Additionally, the SSA is greatly improved during the search process when combined with LAHC, resulting in further improved performances. The given research contributes significantly to the field, as it provides a novel way of finding the best CRNNs for predictions of sequence in handwriting recognition that can be replicated in future and various language and script. |
| first_indexed | 2025-11-15T03:55:57Z |
| format | Thesis |
| id | ump-44601 |
| institution | Universiti Malaysia Pahang |
| institution_category | Local University |
| language | English |
| last_indexed | 2025-11-15T03:55:57Z |
| publishDate | 2024 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | ump-446012025-05-30T02:36:50Z http://umpir.ump.edu.my/id/eprint/44601/ A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm Ahmed Ali Mohammed, Al-saffar QA75 Electronic computers. Computer science The processing of automatically generated images of documents, which is a complex and expensive but necessary component of handwriting recognition, and which has always drawn the attention of engineers and scientists. The definition includes processing of an image in which handwritten letters or numbers exist with the term handwriting recognition. The practical applications of handwriting recognition in many of the current real-world applications rely on the processing of sequential texts, where the language must be detected with efficiency, suggesting the development of a flexible handwriting recognition model. The purpose of this research is to finally describe an automated and flexible way of finding the most appropriate CRNN (Convolutional Recurrent Neural Network) architecture for predictive-sequence related tasks. This research present a dynamic configurator of the CRNN (DC-CRNN), geared for sequence learning in the context of handwriting recognition, inspired by bio-inspired approaches. The built DCCRNN is based on the Salp Swarm optimization Algorithm (SSA), a processor that given a particular dataset will find the best CRNN’s structure and hyperparameters. The present research offers a unique hybridization of SSA with the Late Acceptance Hill-Climbing (LAHC) to further strengthen the optimization process. Experiments were performed on two well-known datasets of Arabic and English handwriting, IAM and IFN/ENIT. The empirical results display that the proposed DC-CRNN and its SSA hybridization are capable of autonomously finding and selecting the best CRNN for a specific dataset. The experimental results indicate that the implementation outperformed other classic handcrafted CRNNs in terms of performance, accuracy, and overall quality for predictive sequential handwriting recognition tasks. Additionally, the SSA is greatly improved during the search process when combined with LAHC, resulting in further improved performances. The given research contributes significantly to the field, as it provides a novel way of finding the best CRNNs for predictions of sequence in handwriting recognition that can be replicated in future and various language and script. 2024-07 Thesis NonPeerReviewed pdf en http://umpir.ump.edu.my/id/eprint/44601/1/A%20sequential%20handwriting%20recognition%20model%20based%20on%20a%20dynamically%20configurable%20convolution%20recurrent%20neural%20network%20and%20hybrid%20salp%20swarm%20algorithm.pdf Ahmed Ali Mohammed, Al-saffar (2024) A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm. PhD thesis, Universti Malaysia Pahang Al-Sultan Abdullah (Contributors, Thesis advisor: Suryanti, Awang). |
| spellingShingle | QA75 Electronic computers. Computer science Ahmed Ali Mohammed, Al-saffar A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| title | A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| title_full | A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| title_fullStr | A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| title_full_unstemmed | A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| title_short | A sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| title_sort | sequential handwriting recognition model based on a dynamically configurable convolution recurrent neural network and hybrid salp swarm algorithm |
| topic | QA75 Electronic computers. Computer science |
| url | http://umpir.ump.edu.my/id/eprint/44601/ http://umpir.ump.edu.my/id/eprint/44601/1/A%20sequential%20handwriting%20recognition%20model%20based%20on%20a%20dynamically%20configurable%20convolution%20recurrent%20neural%20network%20and%20hybrid%20salp%20swarm%20algorithm.pdf |