COMPARISON OF DEEP LEARNING ALGORITHMS FOR SIGN LANGUAGE RECOGNITION

Abstract

Communication has an essential impact in facilitating interaction between individuals. It is a crucial and fundamental way of expressing feelings, thoughts, and opinions. The community of deaf people relies on visual communication of information which uses sign language and speechreading. The significant application of sign language is now a vital part of the hearing-impaired culture. Sign language recognition systems implement machine learning techniques to convey the hand pattern movement into an understandable message. This thesis aims to make a comparative study between two deep learning models, more specifically, the Convolutional Neural Network (CNN) architecture used as feature extractor and classifier and the hybrid model CNN – Support Vector Machine (SVM), which uses the CNN model as feature extractor and SVM algorithm for the classification process. The paper is divided into two parts, the first one lays into a comprehensive study of both models' development, architecture, and design. The second part is about the practical comparison of methods using coding to observe their performance. The methodology used to conduct this study is a combination of literature review and practical application of two used models in data classification and prediction tasks. The techniques used for this project include both the qualitative approach which is used in the first section and the quantitative approach employed in the other section. The thesis dives deeply into the architecture of the models to ensure that each model will perform at maximum capacity so the comparison will be held under the same environment and restrictions. A real-world dataset is taken under consideration to validate the performance of each of the used models. In conclusion, we emphasize the importance of using machine-learning techniques to enhance the interaction of deaf people within society, as well as the efficiency of the model that may be applied for other data classification and prediction tasks.