Hand sign recognition system based on SOM-Hebb hybrid network

This paper discusses a real time Japanese hand sign recognition system. The system uses a hybrid network as a vector classifier. The hybrid network consists of a self-organizing map and a Hebbian learning network. The real time operation makes it possible for users to verify and adjust their input image for the correct recognition. The experimental results show that the visual verification improves the recognition rate of the 41 Japanese hand sign by 15%. In addition, the experimental results show that the use of a hand shape data having a large deviation for the training improves the recognition performance.