الفهرس 
Chapter 1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 138 |  |  |
| | | from | 138 | | |
Abstract
Sign Language (SL) is the main language for handicapped and disabled people. Each country has its own SL that is different from other countries. Each sign in a language is represented with variant hand gestures, body movements, and facial expressions. Researchers in this field aim to remove any obstacles that prevent the communication with deaf people by replacing all device-based techniques with vision-based techniques using Deep Learning (DL) techniques. The thesis highlights two main SL processing tasks: Sign Language Recognition (SLR) and Sign Language Identification (SLI). The task of SLI is targeted to identify the signer language, while the former is aimed to translate the signer conversation into tokens (signs). The thesis addresses the most common datasets used in the literature for the two tasks (static and dynamic datasets that are collected from different datasets) with different contents including numerical, alphabets, words, and sentences from different SLs. It also discusses the devices required to build these datasets, as well as the different pre-processing steps applied before training and testing. The thesis compares the different approaches and techniques applied on these datasets. It discusses both the vision-based and the data-gloves-based approaches, aiming to analyze and focus on main methods used in vision-based approaches such as hybrid methods and deep learning algorithms. Furthermore, the survey presents a graphical depiction and a tabular representation of various SLR approaches.
Sign Language Identification (SLI) and Sign Language Recognition (SLR) are critical issue facing the handicapper people. Also, ordinary people stumble while contacting with deaf people of special needs or while communicating with each other’s, as they use Sign Language (SL) without realizing that. Using DL techniques has helped to bridge this gap of SLI. A well-defined dataset is created with different types of sign language alphabets; American Sign Language (ASL), British Sign Language (BSL), and Arabic Sign Language (ArSL). ArSL was used due to the fact that researchers are not usually included in their works. We evaluated and tested the performance of capsule neural network (CapsNet) with the two most well-known deep learning algorithms LeNet and VGG-16 over our dataset. The main purpose of this thesis is to achieve high results over our dataset and identify multiple signs.
Objectives:
1. The thesis focuses on creating a dataset capable of accurately classifying different forms of sign languages, including American Sign Language, British Sign Language, and Arabic Sign Language. Although the scope of the research does not cover all aspects of these sign languages, the primary goal is to identify their unique characteristics and distinguish between the signs used in each language. It is important to note that the intention is not to integrate Arabic Sign Language with the more commonly used American and British Sign Languages, but rather the main motivation behind this thesis is to create a comprehensive dataset capable of accurately classifying different forms of sign languages.
2. The main objective of the thesis is to classify sign language letters and identify their corresponding language types. Due to the need for sign language speakers to communicate with individuals with disabilities in various conferences, a system capable of recognizing many sign languages is required. To achieve this goal, we used deep learning techniques, including LeNet, VGG-16, and CapsNet, which led to good results.
Results:
1. Great results have been achieved compared to other research, as VGG-16 outperformed other methods with a rate of 99.69% on the training dataset and 99.65% on the testing dataset.
2. We obtained lower accuracy in CapsNet and LeNet compared to VGG-16. We achieved 96.54%, 97.45%, and 94.95% on BSL, ASL, and ArSL, respectively, during inference using the LeNet model.
3. We achieved 98.4848%, 98.4286%, and 99.5652% on ArSL, ASL, and BSL respectively while executing them using the CapsNet model. Using VGG-16, we obtained 99.05%, 98.50%, and 99.69% on ArSL, ASL, and BSL respectively.