Emotion Recognition from Facial Expression using CNN

Abstract

In this paper, a time-efficient hybrid design for emotion recognition using facial expression is proposed which uses pre-processing stages and several Convolutional Neural Network (CNN) topologies to improve accuracy and training time. Sadness, happiness, contempt, anger, fear, surprise, and neutral are the seven primary human emotions anticipated. The model will be tested using the MMA Facial Expression database as well as other facial positions. To avoid bias towards a specific group of photos from a database, performance will be evaluated using cross-validation techniques. Proposed system was trained using a huge database consisting of around 35,000 images. Using our personal system, training time for the proposed model was drastically reduced to 30hrs. Finally, a Web application will be developed to make it more user-friendly in real-time.