Image-Based Number Sign Recognition for Ethiopian Sign Language Using Support Vector Machine

Abstract

In recent times, several technologies have been utilized to bridge the communication gap between persons who have hearing or speaking difficulties and those who dont. Human use speech as a means of communication but hearing impaired people are not fortunate to use their voice for communication, but instead they use sign language as a primary means of communication. However, only a few number of people understand sign language. This results in miscommunication in terms of expression and communication between hearing impaired persons and normal people. To reduce this gap in communication, sign language recognition technologies are used as a means to translate expressions from one form of language to another. Therefore, this paper presents a local number sign recognition approach for Ethiopian sign language using Support Vector Machine (SVM) with sign number images as input and local sign numbers as the desired output. The proposed system comprises of three major stages: Image processing, feature extraction, and SVM modeling. Image processing is performed to make the Amharic sign language number feature extraction process simpler. The features are extracted from the shape of the signers hand to represent local number signs in Ethiopian Sign Language. Finally, a SVM classification model was built using a machine training dataset. The SVM multi-class model achieves 99.33% with a test loss accuracy of 9.7% for local number sign recognition.