BhavnaNet: A Deep Convolutional Neural Network for Facial Emotion Recognition

Facial emotions are vital characteristics that humans use to communicate and understand the feelings of others. Facial recognition has been a challenging task for computer algorithms in the past. Computer vision algorithms can leverage various facial features to understand a person’s emotions. Assessing facial patterns and muscular structures of the face features has been a prominent field of interest for computer vision research as it can be easily used to recognize human emotions. In this paper, a complex human emotion recognition problem is solved using the proposed ‘BhavnaNet’ model. The BhavnaNet is built upon a complex deep neural network architecture that extracts complex features from the facial images and uses them to implement emotion recognition at higher efficiency than the baseline approaches. BhavnaNet is trained on more than thirty-two thousand images from the FER dataset separated into six different emotion categories. The testing of the model is done on the FER dataset as well as a custom dataset of faces of real-world people. The model accurately predicts the facial emotions present on all the human faces that were presented to it in a custom dataset. Using BhavnaNet makes emotion recognition strongly impactful in many research-intensive areas like depression detection, violence detection, deep video surveillance, lie detectors, etc.