期刊名称：International Journal of Advances in Soft Computing and Its Applications
印刷版ISSN：2074-8523
出版年度：2019
卷号：11
期号：3
页码：93-107
出版社：International Center for Scientific Research and Studies
摘要：Sign Language recognition is an important tool for the hearing-impaired in order to communicate with both hearing-impaired and hearing individuals. Similarities in finger-spelling sign language are one of the main factors or problems influencing the accuracy of sign language recognition. This research focuses on one-stroke, Thai finger-spelling sign language (TFSL) in methods of feature extraction with the pyramid histogram of oriented gradients (PHOG) and local features, as well as the application of K-Nearest Neighbors (KNN) recognition. We present, herein, a Thai finger-spelling sign language recognition system (TFSLR) used to classify alphabets shown in similar gestures. Five signers postured fifteen Thai alphabet characters, in which images were taken in five repetitions, totaling 375 finger-spelling images. Our experiment utilizes five-fold cross-validation in order to evaluate the projected system effectiveness. Additionally, we compared the results of each experiment involving PHOG with the amalgamation of PHOG and the local features. The results showed that such amalgamation was capable of handling similarly signed characters with an average accuracy of 97.6%.
关键词：Finger;spelling sign language; histogram of oriented gradients; k;nearest neighbors; PHOG;
