Network of binary SVMs for facial expression recognition

Abstract

With the advancement of technology and the impressive focus on artificial intelligence, the communication between humans and computers had become a very important area of interest in the computer vision domain. The Development of an accurate Facial Expression Recognition (FER) system is therefore a must to enhance the humancomputer interaction. In this project we create a facial expression recognition system that will have the least percentage of error. In order to achieve this, a complex model composed of several stages was developed. As stages, the Viola-Jones was used for facial detection. Histogram of Oriented Gradients (HOG) was used for feature extraction and a network of Support Vector Machines (SVMs) for classification. In order to validate results, the accuracy of our designed system was tested. This is done on three famous datasets that were designed for facial expression recognition, the Japanese Female Facial Expressions, Extended Cohn-Kanade Dataset and the Radboud Facial dataset (RaFD). The datasets are to be used for both training and testing the system. After which, optimization methods are to be applied to increase the accuracy as much as possible. The last step in this project is to reproduce the system as a user friendly application that will make it possible to be used by everyone. This step will be done by producing a GUI using matlab that will take from the camera the face of the user and will return instantaneously the facial expression. The following chapters will discuss in-depth details of each part of our model and will enumerate on the results produced by each code.