Tunisian Sign Language Recognition System of Static Two-Handed Asymmetrical Signs using Deep Transfer Learning

Deaf and Hard of Hearing people use Sign Languages in the interaction among themselves and among hearing people. The automatic recognition of Static Two-Handed Asymmetrical signs is a hard operation, since it involves the implementation of complex processing system for providing image perception. In this paper, we produce a dataset of 2000 images containing 12 Two-handed Asymmetrical Tunisian Signs and utilize transfer learning for automatic recognition, achieving 98.29 % Accuracy. The simulations prove that this best Accuracy value is yielded by the Xception model when combined with the Adagrad optimizer, which indicates that our approach achieves high results despite using a small Dataset.