Facial Expression Recognition with Fused Deep and Geometric Features

Abstract

A novel Facial expression recognition(FER) systemproposed in this study uses a combinationof deep and handcrafted geometric facial features to automatically recognise expressions from input images. The system captures the subtle details in the input image by extracting deep features of the image using a Convolution neural network built from scratch. The handcrafted geometric features used in our work carry the spatial information and the relative distances between various facial landmark points like eyebrows, eyesand mouth etc. Finally ,the two different feature vectors obtained aremerged together in order to create the final features set which is fed to a multi-class SVM for classification. The hyper-parameters for the SVMare also optimised to achieve good recognition accuracy. The performance of the system is assessed by performing experiments on two publically available image databases.Experiments reveal that recognition accuracy of 81.39% and 86.01% is achieved for JAFFE and MUG database respectively.