XCSF for Prediction on Emotion Induced by Image Based on Dimensional Theory of Emotion

Abstract

Affective image classification problem is a problem aims on classifying images according to their affective characteristics of inducing human emotions. This paper extends the discrete state classification problem into a continuous function approximation problem by applying the experimental paradigm of dimensional emotion model. The Extended Classifier System for Function Approximation (XCSF) was applied to the problem and the results suggest that it outperforms linear regression (LR) in accomplishing this task. The obtained results also indicate that without using content based features of the images, the effects of individual difference can be relatively small.