Cursive Script Text Recognition in Natural Scene Images: Arabic Text Complexities

ISBN-13: 9789811512964 / Angielski / Twarda / 2020 / 111 str.

Saad Bin Ahmed; Muhammad Imran Razzak; Rubiyah Yusof

Spis treści

Section#1 Introduction and Challenges

Chapter# 1          Foundations of Cursive Scene Text

1.1            Introduction

1.2            What is Cursive script

1.3            Role of Context in Cursive script

1.5        Applications

1.6            Contribution

 

Chapter# 2          Text in Wild and its Challenges

                                2.1         In-built Complexities Relevant to Cursive Scene Text

                             2.2       Scene Text Localization issues

               2.3        Cursive Scene Text Recognition Limitations

Chapter#3           Arabic Scene Text Acquisition and Statistics

3.1         Importance of Dataset Analysis

3.2         Dataset Collection

3.2.1      Multilingual Dataset Generation

3.2.2      English-Arabic Scene text 42k Dataset

3.5         Pre-processing of Acquired Samples

3.6         Generation and Verification of Ground Truth

 Methods and Algorithms

             Chapter#4          Traditional Approaches

4.1.         Methods Designed for Feature Analysis

4.2         Research Methodologies Designed for Cursive Scene Text

                                4.2.1      Importance of Implicit Segmentation

4.3      Role of Explicit Segmentation

                                4.1         Invariance Feature Extraction in Co-occurrence Extremal Regions

              4.2         Window based features

                                              4.4         Linear spatial pyramid

                                              4.4.1      Formulation and Preprocessing

 

Chapter 5#  Deep Learning

                                5.1         Hybrid Deep Learning Model         

              5.2         Deep Convolutional Neural Network

5.3  RNN

                5.3.1         Why LSTM networks suitable for Cursive Scene Text?

                                5.3.2      Importance of Connectionist Temporal Classification (CTC) in LSTM

5.4         Hierarchical Subsampling based Cursive scene Text Recognition

5.5         Transfer Learning

4.11         Summary

 

                Chapter 6#           Progress in Cursive Wild Text Recognition 

                                5.1          Overview of latest trends

                                5.2          Current Status

(Competition)

                Chapter# 7         Open Research issues and Future Direction

                                6.1          Research problems with perspective of state-of-the-art techniques

                                6.2          Future Directions

O autorze

Dr. Saad Bin Ahmed is a lecturer at King Saud bin Abdulaziz University for Health Sciences, Riyadh, Saudi Arabia (KSAU-HS). He is also associated with Center of Artificial Intelligence and Robotics (CAIRO) research lab at the 

Malaysia-Japan International Insitute of Technology (M-JIIT), Universiti Teknologi Malaysia, Kuala Lumpur, Malaysia. He completed his Ph.D. in Intelligent Systems at the Universiti Teknologi Malaysia in 2019. Before that, he completed his Master of Computer Science in Intelligent Systems at the Technische Universität, Kaiserslautern, Germany, and was a research assistant at the Image Understanding and Pattern Recognition Research Group at the same university. His areas of interests are document image analysis, machine learning, computer vision, and optical character recognition. He has authored more than 25 research articles in leading journals and conferences, as well as book chapters.

 

Dr. Muhammad Imran Razzak is associated with the University of Technology Sydney, Australia. Previously, he was an Associate Professor at King Saud bin Abdulaziz University for Health Sciences. He holds a patent and is also the author of more than 70 papers in respected journals and conferences. He has secured research grants of more than $1.3 million, and has successfully developed and delivered several research projects. His areas of research include machine learning, document image analysis, and health informatics. 

 

Prof. Dr. Rubiyah Yusof is a director at (CAIRO) M-JIIT, Universiti Teknologi Malaysia, Kuala Lumpur, Malaysia. She received her master’s degree in Control Systems from Cranfield Institute of Technology, United Kingdom, in 1986, and her Ph.D. in Control Systems from the University of Tokushima, Japan, in 1994. Throughout her career, Dr. Yusof has made significant contributions to artificial intelligence, process control, and instrumentation design.

She is recognized for her work in biometrics systems, such as KenalMuka (face recognition system) and a signature verification system, which won both national and international awards. She is the author of the book Neuro-Control and its Applications published by Springer-Verlag, in 1995, which was translated to Russian in 2001. Professor Dr Yusof is a member of the AI Society Malaysia, Instrumentation and Control Society Malaysia, and Institute of Electrical and Electronics Engineers Malaysia.

Blurb

This book offers a broad and structured overview of the state-of-the-art methods that could be applied for context-dependent languages like Arabic. It also provides guidelines on how to deal with Arabic scene data that appeared in an uncontrolled environment impacted by different font size, font styles, image resolution, and opacity of text.

 

Being an intrinsic script, Arabic and Arabic-like languages attract attention from research community. There are a number of challenges associated with the detection and recognition of Arabic text from natural images. This book discusses these challenges and open problems and also provides insights into the complexities and issues that researchers encounter in the context of Arabic or Arabic-like text recognition in natural and document images. It sheds light on fundamental questions, such as a) How the complexity of Arabic as a cursive scripts can be demonstrated b) What the structure of Arabic text is and how to consider the features from a given text and c) What guidelines should be followed to address the context learning ability of classifiers existing in machine learning.