American Sign Language Character Recognition with Capsule Networks

Abstract

Sign language is an argument that has vital importance in the life of hard of hearing people for communication with others. With sign language, a person with disabilities can convey his/her opinions and thoughts by using gestures and mimics to another person. Today, understanding and using sign language of people without disabilities along with the people with disabilities is really important to socialize them. With the development of technology, software tools become usable for identification of sign language and various software products are produced. In this study, recognition of sign language characters via a system, which was trained using images of letters in American sign language, is aimed. Capsule networks, which is proposed in the recent period, are used for training and test processes and compared with LeNet which is one of the first successful and currently using a model of deep learning. As a result of the study, it has been seen that capsule networks are useful for sign language character recognition and produced more successful results than LeNet.