Classification of Perceptual Interpretation of Bistable Images Using EEG and Artificial Neural Networks

This paper proposes the use of artificial neuron networks for classification and automatic recognition of human brain states associated with the perception of ambiguous images. Based on the experimental data obtained, the ANN architecture was optimized to achieve up to 95% accuracy in the classification of brain wave patterns during the perception of ambiguous images. We also found features typical for all subjects in the EEG patterns corresponding to different interpretations of the Necker cube, so that a single ANN trained on a single person’s EEG data set can classify the corresponding brain state of a large group of people with high quality. Two sets of experiments with and without key presses demonstrated that motor activity (real or imagined) had no effect on cube classification results.