Recognizing vietnamese sign language based on rank matrix and alphabetic rules

Sign language plays an important role in communication in hard-of-hearing community. Hand gesture recognition is an issue which is being researched widely. In this paper, we propose an approach, which can perform in real-time, to solve such problem for Vietnamese sign language. Instead of RGB data as many other solutions, the input of our system is depth images captured by Microsoft Kinect. We also propose a novel technique, called rank-order correlation matrix (ROCM), to describe hand gestures. Based on properties of Vietnamese alphabet and the captured gesture, the classification stage is applied on different sets of gestures. Multiple support vector machines (SVMs) is combined with "max-wins" voting strategy to perform the recognition task. Experiments are conducted on three datasets of the D-VSL database and receive promising accuracy.