Screen text character selection for people with oral communication problems using EEG signal processing
This paper presents research work related to EEG signal processing in the field of linear feature extraction for people with severe oral communication problems. A battery of characters is shown on a PC screen, one of which is highlighted at 5.7 Hz. When the user is focusing on the target character, his brain generates the P3b component of the P300 Event Related Potentials (ERP). Fisher’s Linear Discriminant Analysis (FLDA) is used as the classification algorithm after a pre-processing stage consisting of filtering, down sampling and channel selection. The database is the “Data Set II” of the BCI Competition III. Using 4 iterations, 80.65% of accuracy is obtained. 90.3225807% of accuracy is achieved using 14 iterations. These results are very useful because we just use a 4 channel system instead of the traditional 8 to 32 channels system where the best results only reach 85%. Future work will combine the presented work with the neural networks classification technique.