EEG and the human perception of video quality: Impact of channel selection on discrimination

The ultimate goal of our research project is to quantify the human perception of video quality directly from brain responses. Specifically, subjects watch a set of video sequences whose qualities vary with time while their brain responses are monitored using a 128-channel electroencephalograph (EEG). We compare four potential sets of time-localized feature vectors-autoregressive coefficients, average time-frequency power, Cepstral coefficients, and entropy-using optimal discriminant formation and LDA-based classification. Of particular interest here is determining which subsets of electrodes provide the most useful information with respect to the different possible sets of feature vectors for discrimination and, ultimately, classification.