A NOVE FACIAL EXPRESSION RECOGNITION SYSTEM USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

The more significant, effective and non-verbal method of communication is the Facial expression using the emotions. In the machine learning field, Facial Expression Recognition (FER) is an important subject. Particularly, Facial recognition system has become the foremost significant research topic in biometrics. The training of system with a small dataset using the available data has a significant impact which can negatively affects the performance. Hence a novel facial expression recognition system using convolutional neural network (CNN) is proposed in this paper. Here, a novel method of extracting geometric features that uses simple techniques for calculating the facial components to obtain the robustness of any pose variation is employed. The huge amount of data is trained in this FER system by using a deep learning architecture of Convolutional Neural Network (CNN) model. Various types of filtering techniques have been applied for augmenting the little amount of small business employee images.  An attempt can also be made to determine which data augmentation options had the greatest impact on facial recognition, so facial recognition by training each image with many features on a new augmented data set is performed with non-real time application. The proposed facial expression recognition system is validated with these obtained features using the augmented emotion data set of Cohn Kanade + (CK+) and provides the six types of emotion results with high accuracy than all of the other FER system technologies.