EEG-based brain computer interface for word recognition using hidden markov models

Abstract

Brain-computer interfaces (BCI) can give augmentative communication and control pathways not only to patients with neuromuscular impairments but also to normal persons. Recently many researchers have developed several BCI systems which can decode user’s desires from EEG. In this thesis, we proposed a new BCI system that can interpret user-intended English words from EEG recorded when he imagined a series of characters of words. Our system is made up of the artificial neural network and the hidden Markov model. It starts from the classification of characters by the former to the recognition of intended words by the latter. It makes use of the temporal fluctuation of EEG signal and the characteristics of words in a dictionary such as Brown Corpus 2000. The result of the proposed system is compared with the previous BCI keyboard systems in the viewpoint of information transfer rate. Our system can transmit information at the rate of 23.8 bits per minute that outperforms the previous systems. And as a specific application that our system can be applied to, a situation that a patient in a hospital uses our system is introduced.