Abstract: This work aims to recognize the main 7 facial emotions (which is quite hard for a variety of reasons) utilizing a novel convolutional neural network (CNN). Moreover, it aims to clearly depict how a convolutional neural network is structured. In order to optimize the network and performance and to deal with things concerned with recognizing emotions, we utilized a variety of techniques, such as max pooling, the activation procedure, the optimizer, and eliminating the overfitting. Additionally, a variety of experiments were conducted utilizing different numbers of layers as well as epochs. Besides, preprocessing and augmentation on the dataset were utilized. To evaluate the performance of the proposed work, tests are conducted on the benchmark dataset, extended Cohn-Kanade, focusing on the key 7 facial expressions. We compared the performance of this work with that of other models conducted on the same dataset. The proposed work outperformed other models.

Keywords: convolutional neural networks, facial emotions, facial emotions recognition, over-fitting, Epoch, data augmentation

Pages: 169-178