Feature Based Myanmar Fingerspelling Image Classification Using SIFT, SURF and BRIEF

Abstract

Deaf people use Sign Language and Fingerspelling as a fundamental communication method. Fingerspelling or manual spelling is a method of spelling words using hand movements, and most often used to spell out names of people, places, organizations, books and other words for which no sign exists. In this experiment, the images for 31 static fingerspelling characters of Myanmar consonant are used as the input images. Three feature vectors extraction methods (SIFT, SURF, and BRIEF) were done separately on our collected Myanmar Sign Language (MSL) fingerspelling images. MSL fingerspelling data are classified with seven different approaches; Multilayer Perceptron, Gaussian Naïve Bays, Decision Tree, Logistic Regression, Random Forest, Support Vector Machine and K-Nearest Neighbor. In this paper, we provide the performance results of different features on different classifiers and the highest classification rate is up to 97% with SURF feature and Random Forest classifier. Moreover, 10- fold cross validation was made in our experiment and we provide the classification results for each classifier.