Facial Emotion Recognition Using Attention Mechanism

Abstract

The proposed project aims to create a facial emotion recognition system using Convolutional Neural Network (CNN) based on the Visual Geometry Group – 16 (VGG-16) classification model. To improve the efficiency of the processing and for higher accuracy results, the attention mechanism will be used for classification as it has an important role in the deep learning field. For comparison, the classification is done on the publicly available FER-2013 dataset, which contains approximately 30,000 facial RGB 48x48 pixel grayscale images of faces with different 7 expressions, as well as the CK+ dataset, which contains 593 image sequences of 123 subjects with a size of 48x48 and consists of seven expressions. Each stage of the model's implementation is meticulously detailed. Various activation functions have been implemented. Experiments on the datasets Fer 2013 and CK+ yielded 67.14 percent and 98.46 percent, respectively, after several experiments on the proposed model. When the Leaky ReLU activation function was utilized in the attention model with the CK+ dataset, the greatest accuracy was 100 percent.