Please use this identifier to cite or link to this item: https://dspace.univ-ouargla.dz/jspui/handle/123456789/30605
Title: Arabic calligraphy style identification
Authors: Mohamed Lamine Kherfi
Belal Khaldi
Kaoudja, Zineb
Keywords: Document Analysis Systems (DAS)
Arabic calligraphy style
Feature extraction
Machine learning
Arabic calligraphy style recognition
Issue Date: 2022
Publisher: University of Kasdi Merbah Ouargla
Series/Report no.: 2022;
Abstract: Calligraphy is the most highly regarded and most fundamental element of Islamic art; and its strong relation to Islamic history. Defining the style could help to determine the period or the region where the text was written. Calligraphy text presents a group of challenges: letter shape is context-sensitive, orthography is very complex, text calligraphy is very related to the used style, and the calligrapher personal touch. This dissertation aims at contributing to the current research in the field of Arabic Calligraphy Style Recognition (ACSR) by developing novel techniques to analyze and to improve the performance of ACSR systems, with a focus on handwritten Arabic Calligraphy texts. The contribution of the present dissertation is fourfold: (1) we collected and made available a dataset of images for Arabic handwritten calligraphy containing 1685 text (line\ sentence). This dataset is freely available for the scientific community; (2) we investigate the impact of combining classifiers on the task of ACSR; we use the Local Phase Quantization (LPQ) descriptor for Arabic calligraphy feature extraction and three different classifiers namely a Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and a Multi-Layer Perceptron (MLP) for style identification; and we compare between four different classifier combination techniques.The observed Arabic calligraphy style recognition rates are in the range from 93.7 % to 94.6 % for the individual classifier and respectively from 93.8 % to 96.5% for classifier decision combination. (3) We study the effect of using different texture descriptors for feature extraction, separately. For text style recognition, we use five different classifiers. with an accuracy rate in the range from 60.4% to 89.8% for RF classifier and respectively from 87.3% to 94.7% for SVM classifier. We find that the pattern-based descriptor with SVM machine learning is the best combination with classification rate 94.7%. (4) we introduce a new descriptor inspired by the Arabic Calligraphy styles: we find that eachstyle has its distinguishing characteristics, so we propose a new computational method to extract these Arabic Calligraphy characteristics, and we use SVM machine learningwith a classifier combination technique for text style identification. The mean recognition rate was from 96.8% to 97.8%.
Description: Computer Science and Information Technology
URI: https://dspace.univ-ouargla.dz/jspui/handle/123456789/30605
Appears in Collections:Département d'informatique et technologie de l'information - Doctorat

Files in This Item:
File Description SizeFormat 
Zineb_KAOUDJA_Doctorat.pdf1,28 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.