A study on the effects of traditional and olfaction enhanced multimedia on pleasantness classification based on brain activity analysis.

Human emotions are recognized in response to content engaging one (audio music) or two human senses (videos). An enhanced sensation with a more realistic feel could be achievable by engaging more than two human senses. In this study, olfaction enhanced multimedia content is generated by synchronizing traditional multimedia content with an olfaction dispenser for engaging olfactory sense in addition to vision and auditory senses. Brain activity of 20 participants (10 males and 10 females) is recorded with a commercially available EEG headband, while engaging with traditional and olfaction enhanced multimedia content. The human brain activity is used to analyze and differentiate the content engaging two (traditional multimedia content) or more than two (olfaction enhanced multimedia content) human senses. For brain activity analysis, we apply a t-test on the power spectra of five frequency sub-bands (delta, theta, alpha, beta, and gamma) of the acquired EEG data in response to traditional and olfaction enhanced multimedia. We observe that alpha, theta, and delta bands are significant in discriminating the response to traditional and olfaction enhanced multimedia content. High brain activity is observed in alpha, theta, and delta bands of frontal channels, while experiencing the olfaction enhanced multimedia content. A user-independent pleasantness classification based on human brain activity is also presented, where classification performance is measured using 10-fold cross validation. We extract features in frequency domain i.e., rational asymmetry (RASM) and differential asymmetry (DASM) from five EEG bands to classify two pleasantness states based on their valence scores using support vector machine (SVM) classifier. Features are further selected based on EEG electrode pair positions and sub-bands. We observed that RASM and DASM features selected from delta band (olfaction enhanced content), and alpha or gamma bands (traditional multimedia content) gives best classification accuracy. We achieved an accuracy of 75%, sensitivity of 77.7%, and specificity of 72.7% in response to olfaction enhanced multimedia content and an accuracy of 68.7%, sensitivity of 71.4%, and specificity of 69.2% in response to traditional multimedia content in classifying pleasant and unpleasant states using SVM. We observed that classification of pleasant state was comparatively better with olfaction enhanced multimedia content than traditional multimedia content.