Determining the effect of emotional images brightness on EEG signals by classification algorithms

This study demonstrated the effects of image brightness on brain activity in neuroscience research, in which the brightness of emotional images had not been taken into account. Electroencephalography recordings from 14 electrode sites of 31 healthy participants were examined during the presentation of original and bright versions of neutral, pleasant and unpleasant images. Power spectra of the recordings were obtained using the short time Fourier transform. The features were extracted from the power spectra for specific time–frequency windows and data obtained from features were classified using support vector machine (SVM), partial least squares regression (PLSR) and k-nearest neighbor (k-NN) algorithms between the original and bright groups for three emotional contents. New features were created with feature combinations providing high classification accuracy. The data obtained from new features were reclassified using SVM, PLSR, k-NN and voting methods between the original and bright groups for three emotional contents. The classification results revealed that the datasets obtained for the original and bright versions of neutral, pleasant and unpleasant images could be separated with 71–81% accuracy. The brightness effect occurred predominantly in the frontal and central regions. This effect was observed in the early time window of visual processing for pleasant and unpleasant images, and in the late time window for neutral images. The findings emphasize that image brightness of affects the power of brain activity and therefore, is an important parameter to be considered in neuroscience research.