Mood analysis using neural networks

Abstract

Identifying mood of a person using the facial expression has been an area of research interest. The research work described in this paper aims to propose a methodology to do the same. In this work, the input from a webcam is fed to the system which is captured frame by frame. The images are preprocessed and a model based on Convolutional Neural Network (CNN) has been used to predict the emotion of a person. The CNN model is trained using the FER2013 dataset to extract features for identification of emotions. The work flow consists of procuring model for training, cleaning the data, building the model and fine-tuning hyperparameters for an optimal level of accuracy. Parallel to this, the functions have been created to take input from webcam and to process the input to feed the model. The final step has been to integrate all the units. The testing of the application has been done for each function separately, then for the entire work as a whole.