An Effective Facial Expression Recognition System

Abstract

This paper propounds an automated Facial Expression Recognition (FER) system based on Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM). The proposed method recognizes facial expressions for static images and real-time applications. Firstly, face and facial parts are detected from the input image. Preprocessing is performed over the detected parts using histogram equalization and image sharpening technique to reduce the illumination effect. Then HOG is used to extract the distinctive image features from facial regions and concatenated into a single feature vector. Finally, the SVM is used for expression classification with a polynomial kernel function. The proposed system is evaluated on renowned JAFFE and Cohn-Kanade facial expression database for seven basic facial expressions. It reveals that the proposed FER system yields an accuracy up to 97.62% and 98.61% for JAFFE and Cohn-Kanade databases, respectively.