Bengali Sign Language Recognition Using Deep Convolutional Neural Network

Abstract

In this paper, we propose a new method for Bengali Sign Language Recognition using Deep Convolutional Neural Networks (DCNN). The method is built to recognize static hand signs of 37 letters of the Bengali alphabet. Conducting tests on three sets of 37 signs (we have used 31 sets of images for 37 different signs) with total 1147 images, with varying the precision of feature points taken on each test. With the use of deep convolutional neural network and utilizing the learned features from a pre-trained network and fine-tuning the top layers of this network, we have achieved a high overall recognition rate of 96.33% on the training dataset and 84.68% on the validation dataset.