A DCNN Based Real-Time Authentication System Using Facial Emotions

Abstract

The purpose of this work is to design a real-time authentication system using Deep Convolutional Neural Networks (DCNN) based on facial emotions that could be used for security checks in public places. The design of the system is conducted in five stages. Initially, facial expression images for the seven categories are captured, preprocessed and loaded into the system, so that there are 10 images for each of the category namely angry, disgust, sad, fear, neutral, happy and surprise. The network architecture is then defined by arranging various deep learning layers sequentially. A fully connected layer is then defined for training followed by a SoftMax layer. The training process is executed followed by classification of a facial features into their appropriate label or category. When the training data of same person was used a F1 score of 70.71 was obtained. When the training set from two persons, real time classification resulted in a F1 score of 85.6. The results reveal that the accuracy can be increased if the training dataset was increased. The limitation was that raw data was used and can be improved by using data augmentation techniques to remove redundant data. The harmonic mean of real-time facial expression classification using the proposed word was considerably high and is suitable for applications like classification and verification.