Facial Emotion Recognition and Detection Using Convolutional Neural Networks with Low Computation Cost

Abstract

People have always had the natural capacity to read emotions from facial expressions. On the other hand, creating a machine that can recognise emotions is a very difficult task. Recent developments in computer vision and machine learning have made it easier to identify emotions in images. This research suggests a convolutional neural network based approach for facial emotion recognition. There are two stages to the Facial Emotion Recognition (FER) system, and Convolutional Neural Networks (CNNs) are used in both. The model is trained during the first phase, and its accuracy is tested during the second phase. The FER model is capable of accurately identifying seven different facial emotions. Our method purposefully sacrifices accuracy % in order to improve computational efficiency and real-time prediction capabilities, whereas previous research primarily focused on accuracy without taking prediction time or processing resources into account. Our suggested approach achieved 71.61% accuracy on the FER2013 dataset, which is marginally less than the accuracy of the most advanced algorithms, which was 75.2%. On the same system, those algorithms took 93 minutes for computation training; our algorithm, however, drastically cut that time down to 58 minutes. We expect that FER's emotion recognition capabilities will be useful in a variety of domains, including robotics, the detection of lies, and the prediction of student behaviour.