Character Recognition using Brain Computer Interface

Abstract

Brain - computer interface (BCI) enables communication between the user and the computer directly using the electrical signals generated in the brain. It usually involves recording of the electroencephalogram (EEG) of the subject when he/she is performing a mental task assigned by the experimenter. The relevant information is then extracted from the EEG signal and the desired output is achieved.
BCI offers people suffering from serious brain diseases like amyotrophic lateral sclerosis (ALS) or paralysis a way to express themselves and communicate with the society. A P300 signal based speller allows the subject to convey a character by simply concentrating on its position on the screen.
P300 signal is a type of an event related potential (ERP) which arises as a response to a less probable event in a train of normal events. To generate a visually evoked P300 signal experimenters use special models called oddball paradigms. In this case, the subject is asked to concentrate on a character in a 6x6 grid of characters displayed on the screen. The rows and columns of the grid are then flashed randomly. The subject is instructed to count the number of times a particular character gets flashed. The EEG signal is acquired using 8 channels during the whole process. An ensemble of Support Vector Machines (ESVM) is used to correctly classify the signals and arrive at the character that the subject was concentrating on.
The accuracy of the ESVM classifier was tested on both the data acquired in the laboratory and the data set II of the third BCI Competition. The results were contrasted to determine the integrity of the acquired data.