A Thermal Imaging Based Classification of Affective States Using Multiclass SVM

Abstract

In research performances, affective computing has become a developing area because of its large use of application in interface of human computer. Recognition of emotion is one of the art techniques state in determining present human being psychological state. Assessment of the emotional state of humans has been traditionally learned using several direct psychological self-reports and psychological measures. There are various measures to recognize emotional states of human such as facial pictures, gestures, neuro-imaging methods and physiological signals. Therefore, some of these approaches need expensive and sizeable equipment which might hinder free motion. Emotions of human are very overlapping in nature and thus it requires an efficient feature-classifier and extractor assembly. It is a novel non-invasive technique to divide emotion of human through thermal face pictures. Invariants of Hu’s moment of different patches have been fused with statistical characteristic of histogram and used as features of robust in machine of multiclass support vector based division. Here 200 highly expressive thermal images are considered for training and 120 images for testing from IVITE database. The proposed system has overall accuracy of 87.50%.