Human Emotion Identification Based on Features Extracted Using CNN

Abstract

Identifying facial expressions is a difficult topic in the classifying of images. Lately, the domain of image recognition, DL (Deep Learning) is gaining more attention. Our work, "a deep Convolution Neural Network" based on VGGNet (Visual Geometry Group network) Sequential-model was proposed for facial emotion identification problem. The model conducted a fixed kernel size and the number of filters was changed in each block. The FER2013 data set was used and it was considered as a challenge set in the field of recognizing human emotions due to the diversity of images in terms of head position, lighting, shadows, age, and gender. The data set was augmented to increase the number of images in each class. For further validation, a set of images from the internet was tested using the proposed model. The model performance obtains an average accuracy of 79% when the augmented FER2013 data set was used. It was proven that the model is effective in facial emotion identification.