Happy and Sad Classification using HOG Feature Descriptor in SVM Model Selection

Abstract

Facial Expression Recognition (FER) of the image is one of the potential research fields. It remains some open problems to be solved such as various head positions, backgrounds, occlusion, face attribute etc., where the FER 2013 dataset give such conditions. In this research, the small balanced dataset used to recognize two common fundamental expression, happy and sad face image as our set conditions. Using SVM as classifier and HOG as feature expression method, this research shows best performance, that is 72% accuracy, in quadratic polynomial kernel with intercept constant $\mathrm{b}=1$ and tolerance constant $\mathrm{C}=0.1$. By using such conditions, minimized pose variant, a conventional approach in FER such SVM and HOG has shown fair performance in the FER 2013 dataset.