Facial Expression Classification using Deep Extreme Inception Networks

Facial expression classification p lays crucial role in human-computer interaction. A large number of automated methods have been proposed since the past decades. Recently, deep learning is broadly applied in computer vision field as well as facial expression classification. The reasons are to avoid complex feature extraction process and obtained satisfied classification performance. In this work, we propose a deep convolutional neural networks (CNNs) model, inspired from XCEPTION, to classify seven groups of facial expressions. To efficiently use of model parameters, the model architecture has only 2.2 million parameters which is about 10 times less than XCEPTION. The experimental results on FER-2013 dataset show that our model offers comparable accuracy (0.7169) to the state-of-the-art methods and the upper-bound level of human accuracy ( 0.65 ± 5). In addition, our model uses less number of parameters than the state-of-the-art models and without using extra features and data augmentation.