Facial Emotion Recognition with InceptionV3 CNN Architecture

The development of deep learning algorithms in recent years has pushed researchers to explore many different topics, one of the most popular being emotion recognition. The emotions that have evolved in the human face have a significant impact on decisions and disputes about a variety of topics. The automatic extraction of these emotions from face photos can aid in human-computer interaction and a variety of other applications. Emotion recognition plays a huge role in many different fields, such as education to analyze a student's facial expression when listening to lectures to help teachers evaluate their teaching method, and even other fields, such as medicine & security, to help analyze irregular behaviors through emotion recognition. By utilizing deep neural networks, the ability to classify patterns and learn complex features is becoming easier. This paper describes how to utilize a deep learning algorithm, the Convolutional Neural Networks (CNN) with InceptionV3 architecture commonly used for Image Recognition to classify seven different emotion classes (angry, afraid, disgusted, happy, neutral, sad, and surprised) with high accuracy. The architecture was chosen instead of other architectures, such as Resnet50 and VGG19, due to the efficiency and less computing power it needs. The Karolinska Directed Emotional Faces (KDEF) dataset was used to train the model which consists of 980 pictures of human faces, each labeled with one of the 7 emotion categories. With fine-tuning of the hyperparameters, the model achieves a maximum accuracy of 91.33%