Facial expression recognition is an emerging research area for improving human and computer interaction. This research plays a significant role in the field of social communication, commercial enterprise, law enforcement, and other computer interactions. In this paper, we propose a time-efficient hybrid design for facial expression recognition, combining image pre-processing steps and different Convolutional Neural Network (CNN) structures providing better accuracy and greatly improved training time. We are predicting seven basic emotions of human faces: sadness, happiness, disgust, anger, fear, surprise and neutral. The model performs well regarding challenging facial expression recognition where the emotion expressed could be one of several due to their quite similar facial characteristics such as anger, disgust, and sadness. The experiment to test the model was conducted across multiple databases and different facial orientations, and to the best of our knowledge, the model provided an accuracy of about 89.58% for KDEF dataset, 100% accuracy for JAFFE dataset and 71.975% accuracy for combined (KDEF + JAFFE + SFEW) dataset across these different scenarios. Performance evaluation was done by cross-validation techniques to avoid bias towards a specific set of images from a database.
Puthanidam, Roshni Velluva, "Time-Efficient Hybrid Approach for Facial Expression Recognition" (2017). Master's Projects. 565.
Available for download on Thursday, December 20, 2018