A Deep Learning Approach for Recognizing Bengali Character Sign Language

Abstract

For many years, researchers are trying to recognize Bengali sign language for helping deafmute people which is very challenging task on the perspective of our country. Every research has its own margins and is still incapable to be used commercially. For that reason, de-vice interpreter is obligate to accommodate that deaf and hard-of-hearing community to communicate with normal people. In this paper, the main target to con-struct a model to recognize Bengali Character Sign Language using deep leaning approach. For that reason, we use Convolutional Neural Network (CNN) to train individualsigns. For those individual signs, we construct a data set called Bengali Ishara-Lipi to achieve our goal. This model is trained by 5760 preprocessed images and tested by 1440 pictures. The quantitative relation of the trained and test-ed pictures was 80% and 20% severely. Finally, our model gained 92.7% accuracy to recognize Bengali alphabetical sign language. Our model will avail for commencing to make Bengali sign language device interpreter.