Identifying Artistic Expertise Difference in Emotion Recognition in Response to Oil Paintings

Previous studies have been conducted on building emotion recognition frameworks and enhancing their performances using Electroencephalography (EEG) and eye tracking signals. However, the differences between experts in art and non-experts in emotion recognition still remain to be elucidated. In this paper, we systematically evaluate the performance of various computational models for emotion recognition in response to oil paintings and identify the differences between experts in art and non-experts. The experimental results demonstrate that Transformer neural networks achieve the highest accuracies of 65.27% in three-category emotion recognition (negative, neutral, and positive) in response to oil paintings. Although the overall emotion recognition accuracies of the two groups are similar, the mean accuracy of the non-expert group for positive emotion is higher than that of experts, and the expert group has higher recognition accuracy in neutral emotion than the non-expert group. We further investigate the neural patterns of the three emotions in the two groups. The experimental results indicate that neural pattern differences do exist in both emotions and artistic expertise. The parietal and occipital lobes are more activated for positive emotion in the artistic expert group in the alpha, beta, and gamma bands. Our proposed methods provide an understanding of underlying emotion-expertise neurological mechanisms and cognitive processes.