Device independence based on BSOFMS in sign language recognition

While many researchers have documented methods for recognizing sign language from instrumented gloves with high accuracy, these systems suffer from notable limitations: device-dependence and lack of extensibility. An ideal recognition system should be able to switch among a variety of input devices without retraining the entire system. A bidirectional self-organizing feature maps (BSOFMs) is presented in this paper to address this problem. BSOFMs has the ability to convert the vector from different input spaces with heterogeneous representation into one in a unique feature space and enable us to get the same description. In the training process, the raw data produced by different gloves work as input and ideal output alternately. Then the device-dependent description of the hand shape is converted to the compact feature output in the device independent feature space. Based on these models, the incorporation of BSOFMs and the existing recognition framework is introduced and it is crucial to creating a useful device-independent system. Experimental results demonstrate that the proposed system has an ideal performance in device-independent application.