Development of Image-Based Emotion Recognition using Convolutional Neural Networks

In recent years, artificial intelligence has been utilized in many applications. One of the prominent applications is detecting emotion from an image, which can help an intelligent automatic response system respond appropriately based on the user’s emotion. This paper presented the development of emotion recognition using Convolutional Neural Networks (CNN) on image input. First, the extended Cohn-Kanade image emotion database was selected with five defined emotions: happy, sad, anger, fear, surprise, and neutral. Second, face detection and facial landmarks extraction was applied to the input image. Then, the AlexNet model is used as the selected deep learning architecture for transfer learning. Results showed that around 98.2% recognition accuracy could be achieved. Furthermore, precision, recall, and F1-score were evaluated, and it showed the effectiveness of our proposed algorithm.