Please use this identifier to cite or link to this item:
https://ptsldigital.ukm.my/jspui/handle/123456789/513441Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | Khairuddin Omar, Prof. Dr | - |
| dc.contributor.author | Alaa Sulaiman (P77885) | - |
| dc.date.accessioned | 2023-10-16T04:36:42Z | - |
| dc.date.available | 2023-10-16T04:36:42Z | - |
| dc.date.issued | 2020-07-06 | - |
| dc.identifier.other | ukmvital:130946 | - |
| dc.identifier.uri | https://ptsldigital.ukm.my/jspui/handle/123456789/513441 | - |
| dc.description | Handwritten document analysis and recognition (HDAR) is one of the challenging and hot topic in pattern recognition. The main goal of HDAR is to extract information from the written passage. These written passages come from different sources such as historical passage, music score, palm leaf., In addition, several challenges such as the condition of the paper, like poor quality, pale text, fade away and ink smear age, illumination, further increases the source variation. Thus, the pre-processing step is required to enhance the readability of these images by removing unwanted information like background, noise and finally converting them to a binary format. Thresholding binarization algorithms are promising methods for image binarization. However, these methods highly rely on their parameters that need to be determined manually. Regardless of the binarization process, feature extraction from handwritten images plays a significant role in HDAR. A desirable feature extraction algorithm should be implemented in a way that to cope with challenges like the variation in handwritten word’s structure and shape information, variable length, noise, and various writing styles. In this respect, several methods based on key point descriptors and text structural information have been proposed in the literature. Although these descriptors provide a length and language independent description, they miss word’s shape and content information and are less effective in describing detail information. This thesis will focus on three main objectives: (1) handwriting image binarization as a pre-processing phase by enhancing Howe binarization method with tuning some parameters which is the range of high Canny edge detection threshold and Sigma values from the original Howe method, then adding post processing stage which is to needed to include the pixels around the text stroke edges and add it to binarized image just to reduce any loss of pixels around the text edges; (2) two stream deep neural network for handwritten word recognition method based on jointly learning deep features from character and word level representation; (3) new data augmentation methods for boosting the word recognition performance will be studied. In which, two different psychological assumption will be considered in generating new samples. The proposed binarization algorithm evaluated on the different versions of DIBCO data set, and a proposed dataset which include Arabic degraded historical manuscript using F-measures, PSNR and DRD and obtained a significant improvement compared to winner algorithms on DIBCO competition. In addition, the proposed handwritten word recognition have applied on several public datasets including IFN/ENIT version 1, 2, Alexuw, CVL, and IAM datasets, using various measurement metrics and achieved superior performance to the recent approaches.,Ph.D | - |
| dc.language.iso | eng | - |
| dc.publisher | UKM, Bangi | - |
| dc.relation | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | - |
| dc.rights | UKM | - |
| dc.subject | Universiti Kebangsaan Malaysia -- Dissertations | - |
| dc.subject | Dissertations, Academic -- Malaysia | - |
| dc.subject | Handwritten document image | - |
| dc.subject | Deep neural networks | - |
| dc.subject | Image processing -- Digital techniques | - |
| dc.title | Handwritten document image analysis using deep neural networks | - |
| dc.type | Theses | - |
| dc.format.pages | 219 | - |
| Appears in Collections: | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| ukmvital_130946+Source01+Source010.PDF Restricted Access | 4.32 MB | Adobe PDF |  View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.