Multilayer method based on multi-resolution feature extracting and MVC dimension reducing method for sign language recognition

Hidden Markov model (HMM) has been successfully used in the sign language recognition (SLR). However, due to large vocabulary of the sign language, traditional one-layer HMM method is becoming limited with the increasing number of training samples. It is tardy when recognizing which cannot meet the real time requirement. In this paper, we present a multi-resolution feature extracting method and a reducing dimension method of maximum variance criterion (MVC), which has better performance in sign language recognition system than traditional reducing dimension methods of PCA or ICA. Our multilayer sign language recognition system increases the recognition accuracy by 3.42%, as well as reduces the recognition time by 0.992 second in average, compared with traditional HMM based system.