An efficient method for sign language recognition from image using convolutional neural network

Abstract

Recognition of the sign language is one of the most important milestone in image recognition field. Such systems can help deaf people to communicate with the world. We feel privileged to present a new method which translates from American Sign Language (ASL) fingerspelling into a letter using Convolutional Neural Network and transfer learning. The method is using Google pre-trained model named MobileNet V1 which was trained on the ImageNet image database. Our model was trained on the dataset from Surrey University. We developed a useful model not only for desktop computers but it is also possible to apply it into mobile systems, because of low memory consumption.