An affective classifi cation approach for detecting emotions evoked in static food images

Abstract

To make machine emotionally intelligent and to have a ective decision making, emotion is one of the basic human attribute which is used in a ective classi cation . In this research work food images are used to evoke emotions by using principle of art features(emphasis, gradation, variety, symmetry, movement and harmony). These features are extracted from the images to form a feature vector and as a result a food-emotion model is created. Emotional labels to these images are associated with help of valence arousal psychological model. Three classi ers SVM(Support Vector Machine), MLP (Multi-layer Perceptron ) and Naive Bayes are used to form a classi cation model. This classi cation approach help in forming a 4-label and 3-label model which associate emotional attribute Happy, Relax,Stress and Boring in the food images. With the help of these results decision making become more related to human attribute of emotion. Among all models, 3-label model give accuracy of 58%.