American Sign Language recognition using Support Vector Machine and Convolutional Neural Network

Abstract

A sign language recognition system is an attempt to help the speech and the hearing-impaired community. The biggest challenge is to recognize a sign accurately. This can be achieved by training the computers to identify the signs. The accuracy depends on the methods used for classification and prediction which is achieved through machine learning. This research proposes the recognition of American Sign Language by using Support Vector Machine (SVM) and Convolutional Neural Network (CNN). In this work we have also calculated optimal filter size for single and double layer Convolutional Neural Network. In the first phase features from the dataset are extracted. After applying various preprocessing techniques, Support Vector Machine with four different kernels i.e., ‘poly’, ‘linear’, ‘rbf’ and ‘sigmoid’ and Convolutional Neural Networks with single and double layer are applied on training dataset to train the model. Finally, accuracy is calculated and compared for both the techniques. In CNN filters of different sizes have been used and optimal filter size has been found. The experimental results establish that the double layer Convolutional Neural Network achieve an accuracy of 98.58%. Optimal filter size is found out to be 8 × 8 for both single and double layer Convolutional Neural Network. From the experimental results we conclude that accuracy of CNN model can be improved by altering the filter size. This also helps in CNN to learn optimum values for variable sized parameters and tuning of different hyper parameters.