An Emotion Estimation Modeling Approach using Graphical Parameters of Affective Pictures

Wasantha Samarathunga, Koji Makino, Hiroshi Hashimoto, and Yasuhiro Ohyama


emotion modeling, affective image classification, neural network, IAPS


We propose affective image classification on dimensional affective groups against conventional discrete affective groups which does not cover all three emotional vectors. This approach serves as affective group identification phase of a full emotion estimation model from affective pictures. The known emotional vector values Valence, Arousal and Dominance of IAPS photo set for all subjects are categorized into nine affective groups using few well known clustering algorithms and the clusters are evaluated by using photo description. The texture and histogram based graphical features of photographs without borders within the SOM based affective groups are evaluated by using the pattern recognition tools of neural network to identify the affective groups where the inputs originated. The test results proved that the gray scale is suitable for high valence or high arousal groups, edges and corners are suitable for high valence groups and blue histogram is useful in high arousal groups. Single or logical combination of these trained networks can be used for affective group recognition in high arousal or valence group. High dominance groups are under development.

Important Links:

Go Back