An Accurate Sign Language Recognizer Using Random Sphere Oracle

In the present study, we propose an automatic sign language recognizer using an ensemble method called Random Sphere Oracle (RSO). The aim of using this method is to improve the accuracy of the existing sign language recognition (SLR) systems by using an efficient classifier to classify the sign language gestures into their respective classes. The design of the classifier composes of three steps. The first step aims to select the base classifier for the ensemble method, and this can be achieved by conducting a comparison between the classification accuracy of five benchmark classifiers, namely, Linear Discriminant Analysis (LDA), Classification and Regression Tree (CART), Naive Bayes (NB), Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs). The best classifier is used in the second step as a base classifier to generate a set of 50 classifiers using RSO. The last step' aim is to select among the 50 classifiers created in the second step, a subset of classifiers with the best performance. For this purpose, we apply a multi-objective genetic algorithm called controlled elitism NSGA-II to search for the best subset of classifiers that maximize two objectives: accuracy and diversity.All the classifiers are applied on a benchmark sign language data set and the results show that the proposed method has successfully improved the accuracy of the SLR system that uses the best single classifier.